Professional Claude Code skills marketplace featuring production-ready skills for enhanced development workflows.
Python
1,458
1,397 commits
updated Oct 14, 2026
Professional Claude Code skills marketplace featuring production-ready skills for enhanced development workflows.
⭐ Start here if you want to create your own skills!
The skill-creator is the meta-skill that enables you to build, validate, and package your own Claude Code skills. It's the most important tool in this marketplace because it empowers you to extend Claude Code with your own specialized workflows.
This is a production-hardened fork of Anthropic's official skill-creator, born from building real skills and hitting every wall the official version doesn't warn you about.
The official skill-creator tells you what to build. Ours also tells you what not to try — and why.
| You're trying to... | Official | This Fork |
|---|---|---|
| Research before building | "Check available MCPs" | Ordered search protocol with decision matrix: Adopt / Extend / Build |
| Create a skill interactively | Prose-based instructions | Structured AskUserQuestion checkpoints — user never loses context |
| Avoid common mistakes | No guidance | Cache edit warnings, prerequisite checks, security scan gate |
| Know the architecture options | Not mentioned | Inline vs Fork decision guide with examples (choosing wrong silently breaks your skill) |
| Validate before shipping | Basic YAML check | Expanded structural validator plus provenance-checked old-vs-new capability audit; packaging re-verifies the completed review instead of trusting a marker |
| Match validation cost to risk | One full create-test-review loop | Risk tier and evaluation spend are separate: targeted checks first; paired benchmarks require explicit authorization or a decision-bearing evidence plan plus opt-in |
| Catch security issues | No tooling | security_scan.py with gitleaks integration — hard gate before packaging |
| Learn from real failures | No failure cases | Battle-tested methodology with documented failure patterns and gotchas |
| Distill past conversations safely | Not covered | Explicit local manifest, message-level time window, redaction, opaque source IDs, ignored .enrich/ staging, and manual promotion into references/scripts |
| Turn approved artifacts into skill behavior | Not covered | Artifact-corpus distillation workflow: script-measured pattern extraction (≥3-artifact threshold), cataloging≠distillation gate, invariant-vs-register boundary, independent completeness audit |
| Survive concurrent sessions on one skill repo | Not covered | git-ref baselines, re-read-before-write, HEAD check before commit, own-paths-only staging, one-bump-per-outcome versioning |
| Ground knowledge skills in evidence | General advice | Authority ladder from real calls and machine-readable specs through production code, plus executable-example smoke checks and evidence-boundary rules |
| Have both installed at once | Coin flip — the two descriptions are near-identical | Detects the clash on trigger and offers a one-command, reversible SessionStart routing hook (only ever installed when both coexist); the official plugin stays usable by explicit request |
| Your own skill collides with an installed plugin | Not covered | generate_supersede_kit.py stamps the same conditional routing kit into your skill, plus a measured precedence decision guide (rename → description tiebreaker → hook → disable) |
Full methodology: skill-creator/references/skill-development-methodology.md
In Claude Code (in-app):
/plugin marketplace add daymade/claude-code-skills
Then:
From your terminal (CLI):
claude plugin marketplace add https://github.com/daymade/claude-code-skills
# Marketplace name: daymade-skills (from marketplace.json)
claude plugin install daymade-skill@daymade-skills
After installing skill-creator, simply ask Claude Code:
"Create a new skill called my-awesome-skill in ~/my-skills"
"Validate my skill at ~/my-skills/my-awesome-skill"
"Package my skill at ~/my-skills/my-awesome-skill for distribution"
Claude Code, with skill-creator loaded, will guide you through the entire skill creation process - from understanding your requirements to packaging the final skill.
📚 Full documentation: daymade-skill/skill-creator/SKILL.md
📝 Initialize New Skill
✅ Validate Skill Structure
📦 Package Skill for Distribution
/plugin marketplace add daymade/claude-code-skills
Then:
macOS/Linux:
curl -fsSL https://raw.githubusercontent.com/daymade/claude-code-skills/main/scripts/install.sh | bash
Windows (PowerShell):
iwr -useb https://raw.githubusercontent.com/daymade/claude-code-skills/main/scripts/install.ps1 | iex
Add the marketplace:
claude plugin marketplace add https://github.com/daymade/claude-code-skills
Marketplace name is daymade-skills (from marketplace.json). Use @daymade-skills when installing plugins.
Do not use the repo path as a marketplace name (e.g. @daymade/claude-code-skills will fail).
In Claude Code, use /plugin ... slash commands. In your terminal, use claude plugin ....
Essential Skill (recommended first install):
# skill-creator ships inside the daymade-skill suite
claude plugin install daymade-skill@daymade-skills
Documentation Suite (shared namespace for document workflows):
claude plugin install daymade-docs@daymade-skills
This suite exposes related skills under one namespace, including:
/daymade-docs:docs-router
/daymade-docs:doc-to-markdown
/daymade-docs:mermaid-tools
/daymade-docs:pdf-creator
/daymade-docs:ppt-creator
/daymade-docs:docs-cleaner
/daymade-docs:excel-automation
/daymade-docs:read-docx-review
These skills ship as a bundle — there are no separate single-skill plugins. All documentation skills live under daymade-docs/ and install together from the suite.
The router handles automatic selection. The specialist commands remain available for manual use;
ppt-creator is manual-only. New presentation creation uses deck-creator when installed.
Apple Platform Suite (shared namespace for macOS and iOS development/operations):
claude plugin install daymade-macos@daymade-skills
/daymade-macos:capture-screen
/daymade-macos:developing-ios-apps
/daymade-macos:macos-cleaner
/daymade-macos:macos-load-doctor
/daymade-macos:macos-permissions
/daymade-macos:macos-watchdog
These skills are bundle-only under daymade-macos.
Codex Suite (shared namespace for Codex workstation setup, assisted coding, and visual exploration):
claude plugin install daymade-codex@daymade-skills
Codex CLI and Desktop users can install the same suite through Codex's plugin marketplace:
codex plugin marketplace add daymade/claude-code-skills
codex plugin add daymade-codex@daymade-skills
/daymade-codex:codex-image-gallery
/daymade-codex:local-codex
/daymade-codex:design-style-picker
/daymade-codex:interaction-design-board
/daymade-codex:codex-1m-context-window-setup
These skills are bundle-only under daymade-codex.
Claude Code Operations Suite (shared namespace for Claude Code power-user workflows):
claude plugin install daymade-claude-code@daymade-skills
This suite bundles the skills that extend Claude Code itself — cross-project prior-work retrieval across code, docs, Skills, meetings, WeChat archives, and conversation history; fast local conversation discovery across Claude Code and Codex; session recovery; CLAUDE.md tuning; version-synced Lark CLI routing; troubleshooting; statusline configuration; export repair; marketplace development and suite consolidation; terminal screenshot rendering; usage analysis; and multi-provider model switching:
/daymade-claude-code:claude-code-ops-router
/daymade-claude-code:local-conversation-history
/daymade-claude-code:read-claude-code-history
/daymade-claude-code:read-codex-history
/daymade-claude-code:continue-claude-code-work
/daymade-claude-code:continue-codex-work
/daymade-claude-code:claude-skills-troubleshooting
/daymade-claude-code:claude-md-progressive-disclosurer
/daymade-claude-code:statusline-generator
/daymade-claude-code:fixing-claude-export-conversations
/daymade-claude-code:marketplace-dev
/daymade-claude-code:terminal-screenshot
/daymade-claude-code:claude-usage-analyst
/daymade-claude-code:claude-switch-models-setup
/daymade-claude-code:read-claude-web-conversation
/daymade-claude-code:claude-migrate-memory-to-doc
/daymade-claude-code:claude-code-hooks
/daymade-claude-code:prior-work-retrieval
/daymade-claude-code:lark-cli-router
/daymade-claude-code:claude-code-ping-start-5h-quota
/daymade-claude-code:agent-web-search-setup
/daymade-claude-code:tech-selection
Installed names render as daymade-claude-code:<skill> under a single shared namespace. These skills are bundle-only — install the suite to get all members.
Financial Data Suite (shared namespace for investment-research and financial data workflows):
claude plugin install daymade-financial@daymade-skills
This suite bundles the skills that fetch and analyze financial data — Bigdata.com (RavenPack) structured financials and sentiment, US equity fundamentals via yfinance, Gangtise (岗底斯) OpenAPI research suite orchestration, A-share news and policy aggregation, A-share pharmaceutical sector daily reporting, structured devil's-advocate pressure-testing of investment theses, and adversarial due diligence on inflated benchmark claims:
/daymade-financial:financial-router
/daymade-financial:bigdata-skill
/daymade-financial:financial-data-collector
/daymade-financial:gangtise-copilot
/daymade-financial:ashare-news-fetcher
/daymade-financial:daymade-sector-research
/daymade-financial:pharma-daily-report
/daymade-financial:devils-advocate
/daymade-financial:benchmark-due-diligence
Installed names render as daymade-financial:<skill> under a single shared namespace. These skills are bundle-only — install the suite to get all members.
Install Other Skills:
# Research-driven improvement of a specified existing Skill
claude plugin install skill-evolver@daymade-skills
# GitHub operations
claude plugin install github-ops@daymade-skills
# Teams communication
claude plugin install teams-channel-post-writer@daymade-skills
# Local Claude/Codex agent messaging
claude plugin install peer-message@daymade-skills
# Ghostty terminal session snapshot/restore
claude plugin install ghostty-use@daymade-skills
# Repomix extraction
claude plugin install repomix-unmixer@daymade-skills
# AI/LLM icons
claude plugin install llm-icon-finder@daymade-skills
# CLI demo generation
claude plugin install cli-demo-generator@daymade-skills
# Cloudflare diagnostics
claude plugin install cloudflare-troubleshooting@daymade-skills
# UI design system extraction
claude plugin install ui-designer@daymade-skills
# YouTube video/audio downloading
claude plugin install youtube-downloader@daymade-skills
# Secure repomix packaging
claude plugin install repomix-safe-mixer@daymade-skills
# Full audio suite (ASR + transcript correction + meeting minutes + TTS)
claude plugin install daymade-audio@daymade-skills
# Video comparison and quality analysis
claude plugin install video-comparer@daymade-skills
# QA testing infrastructure with autonomous execution
claude plugin install qa-expert@daymade-skills
# Prompt optimization using EARS methodology
claude plugin install prompt-optimizer@daymade-skills
# CCPM skill registry search and management
claude plugin install daymade-skill@daymade-skills
# Promptfoo LLM evaluation framework
claude plugin install promptfoo-evaluation@daymade-skills
# Twitter/X content fetching
claude plugin install twitter-reader@daymade-skills
# Skill quality review and improvement
claude plugin install daymade-skill@daymade-skills
# GitHub contribution strategy
claude plugin install github-contributor@daymade-skills
# Windows Remote Desktop / AVD connection diagnosis
claude plugin install windows-remote-desktop-connection-doctor@daymade-skills
# Product analysis and optimization
claude plugin install product-analysis@daymade-skills
# Scrapling CLI extraction and troubleshooting
claude plugin install scrapling-skill@daymade-skills
# Tencent IMA knowledge base companion and installer
claude plugin install ima-copilot@daymade-skills
# Export Douban (豆瓣) book/movie/music/game collections to CSV
claude plugin install douban-skill@daymade-skills
# Terraform operational traps and multi-environment reliability patterns
claude plugin install terraform-skill@daymade-skills
# Evaluate any LLM endpoint across speed, concurrency, protocol, and quality
claude plugin install llm-eval-harness@daymade-skills
# Video/GIF memes with motion-tracked image overlays
claude plugin install meme-creator@daymade-skills
Standalone plugins can be installed independently; suite members install together with their suite.
For Chinese users: We highly recommend using CC-Switch to manage Claude Code API provider configurations.
CC-Switch enables you to:
Setup: Download from Releases, install, add your API configs, and switch via UI or system tray.
For full documentation in Chinese, see README.zh-CN.md.
GitHub operations through gh CLI and GitHub APIs, with explicit targets, impact previews, and independent state verification.
When to use:
Key features:
🎬 Live Demo
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:docs-router)
Selects the bundled specialist for document conversion, Word and PDF production,
Mermaid images, macOS Excel automation, scanned PDFs, DOCX review extraction, or
documentation cleanup. It reads the selected specialist's full instructions and
required references. Specialist slash commands remain available for manual use.
New presentation creation uses deck-creator when installed; ppt-creator is manual-only.
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:doc-to-markdown)
Converts documents to markdown with Windows/WSL path handling and PDF image extraction.
When to use:
Key features:
🎬 Live Demo
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:mermaid-tools)
Extracts Mermaid diagrams from markdown and generates high-quality PNG images.
When to use:
Key features:
🎬 Live Demo
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:claude-code-ops-router)
Routes plugin and Skill repair, marketplace work, statusline, model profiles and source sync, 1M context-window repair, usage and quota timers, memory migration, and exported .txt repair to one bundled specialist. The selected Skill's full instructions are read at use time; its original slash command remains available for manual use.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:statusline-generator)
Configures Claude Code statuslines with multi-line layouts and cost tracking.
When to use:
Key features:
🎬 Live Demo
Creates educational Teams channel posts for internal knowledge sharing.
When to use:
Key features:
🎬 Live Demo
Follow the local coordination workflow for task/resource declarations, native preflight installation, recovery and receiver checks.
An experimental paired-network route adds invitations and cited answers from owner-selected documents through existing Claude/Codex hosts. It requires an operator-provided relay; the public hosted network and user-adoption claims are not part of this preview.
Install:
claude plugin install peer-message@daymade-skills
Follow the routing contract for native communication and transport gaps.
When to use:
crossSessionInbound)📚 Execution entry and detailed SOPs.
Extracts files from repomix-packed repositories and restores directory structures.
When to use:
Key features:
🎬 Live Demo
Access 100+ AI model and LLM provider brand icons from lobe-icons library.
When to use:
Key features:
🎬 Live Demo
Generate professional animated CLI demos and terminal recordings with VHS automation.
When to use:
Key features:
🎬 Live Demo
Investigate and resolve Cloudflare configuration issues using API-driven evidence gathering.
When to use:
Key features:
🎬 Live Demo
Extract design systems from reference UI images and generate implementation-ready design prompts.
When to use:
Key features:
🎬 Live Demo
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:ppt-creator)
Create persuasive, audience-ready slide decks from topics or documents with data-driven charts and dual-format PPTX output.
When to use:
Key features:
🎬 Live Demo
Download YouTube videos and audio using yt-dlp with robust error handling and automatic workarounds for common issues.
For spoken-content text, reuse existing transcripts first, then fetch original subtitles; download audio for ASR only when those checked routes miss. File-download requests keep the existing workflow.
When to use:
Key features:
🎬 Live Demo
Safely package codebases with repomix by automatically detecting and removing hardcoded credentials before packing.
When to use:
Key features:
🎬 Live Demo
Coming soon
Install:
claude plugin install daymade-audio@daymade-skills(suite-only — invoked asdaymade-audio:transcript-fixer)
Correct speech-to-text (ASR/STT) errors with a Stage 1 dictionary pre-filter, a required Native AI whole-transcript pass, and an audio-backed human gate for unresolved wording, names, numbers, and entities.
When to use:
Key features:
Example workflow:
# Initialize and add corrections
uv run scripts/fix_transcription.py --init
uv run scripts/fix_transcription.py --add "错误词" "正确词" --domain general
# Run full correction pipeline
uv run scripts/fix_transcription.py --input meeting.md --stage 3
# Open only this transcript's unresolved items and listen before deciding
uv run scripts/review-dashboard/server.py --file "/absolute/meeting.md"
# Read the human verdicts back; pending_total must be 0 before final delivery
uv run scripts/fix_transcription.py \
--list-review --review-file "/absolute/meeting.md" \
--review-status all --json
📚 Documentation: See daymade-audio/transcript-fixer/references/ for workflow guides, SQL queries, troubleshooting, best practices, team collaboration, and API setup.
Requirements: Python 3.10+ and uv. Native AI correction uses the active agent; an external API key is needed only for the optional agent-less API route.
Compare two videos and generate interactive HTML reports with quality metrics and frame-by-frame visual comparisons.
When to use:
Key features:
Example usage:
# Basic comparison
python3 scripts/compare.py original.mp4 compressed.mp4
# Custom output and frame interval
python3 scripts/compare.py original.mp4 compressed.mp4 -o report.html --interval 10
# Batch processing
for original in originals/*.mp4; do
compressed="compressed/$(basename "$original")"
output="reports/$(basename "$original" .mp4).html"
python3 scripts/compare.py "$original" "$compressed" -o "$output"
done
🎬 Live Demo
Coming soon
📚 Documentation: See video-comparer/references/ for quality metrics interpretation, FFmpeg commands, and configuration options.
Requirements: Python 3.8+, FFmpeg/FFprobe (install via brew install ffmpeg, apt install ffmpeg, or winget install ffmpeg)
Establish world-class QA testing processes with autonomous LLM execution, Google Testing Standards, and OWASP security best practices.
When to use:
Key features:
Example usage:
# Initialize QA project (creates full infrastructure)
python3 scripts/init_qa_project.py my-app ./
# Calculate quality metrics and gates status
python3 scripts/calculate_metrics.py tests/TEST-EXECUTION-TRACKING.csv
# For autonomous execution, copy master prompt from:
# references/master_qa_prompt.md → paste to LLM → execute the project's approved test plan
🎬 Live Demo
Coming soon
📚 Documentation: See qa-expert/references/ for:
master_qa_prompt.md - Autonomous execution entrygoogle_testing_standards.md - AAA pattern, coverage thresholds, OWASP testingday1_onboarding.md - Onboarding timeline for new QA engineersground_truth_principle.md - Preventing doc/CSV sync issuesllm_prompts_library.md - 30+ ready-to-use QA promptsRequirements: Python 3.8+
💡 Innovation: The autonomous execution capability (via master prompt) enables LLM to execute entire test suites 100x faster than manual execution, with zero human error in tracking. Perfect for third-party QA handoffs - just provide the master prompt and they can start testing immediately.
Transform vague prompts into precise, well-structured specifications using EARS (Easy Approach to Requirements Syntax) - a methodology created by Rolls-Royce for converting natural language into testable requirements.
Methodology inspired by: 阿星AI工作室 (A-Xing AI Studio), which pioneered combining EARS with domain theory grounding for practical prompt enhancement.
When to use:
Key features:
Example usage:
# Before (vague)
"Build me a password reset feature"
# After EARS transformation (7 atomic requirements)
1. When user clicks "Forgot Password", the system shall display email input field
2. When user submits valid email, the system shall send password reset link valid for 1 hour
3. When user clicks reset link, the system shall verify token has not expired
4. When token is valid, the system shall display password creation form requiring minimum 12 characters, 1 uppercase, 1 number, 1 special character
5. When user submits new password meeting requirements, the system shall hash password with bcrypt and invalidate reset token
6. When user attempts password reset more than 3 times in 1 hour, the system shall block further attempts for 1 hour
7. If reset token has expired, the system shall display error message and option to request new link
# Enhanced with domain theories
- Zero Trust Architecture (verify at each step)
- Defense in Depth (rate limiting + token expiration + password complexity)
- Progressive Disclosure (multi-step UX flow)
# Full prompt includes Role, Skills, Workflows, Examples, Formats
🎬 Live Demo
Coming soon
📚 Documentation: See prompt-optimizer/references/ for:
ears_syntax.md - Complete EARS patterns and transformation rulesdomain_theories.md - 40+ theories mapped to domains with selection guidanceexamples.md - Full transformation examples with before/after comparisons💡 Innovation: EARS methodology eliminates ambiguity by forcing explicit conditions, triggers, and measurable criteria. Combined with domain theory grounding (GTD, BJ Fogg, Gestalt, etc.), it transforms "build a todo app" into a complete specification with behavioral psychology principles, UX best practices, and concrete test cases - enabling test-driven development from day one.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:local-conversation-history)
See the Skill's usage and routing contract for history recall, interrupted work and conversation-backup health checks.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:read-claude-code-history)
Read, search, and export Claude Code history across all active config homes and
the long-term archives registered in ~/.claude/history-sources.json, without
resuming or modifying the old task.
When to use:
Key features:
read-codex-history--all-projects searches every Claude project when the project is unknown; --exclude-session skips self-matchesExample usage:
# List recent sessions for a project
python3 scripts/analyze_sessions.py list /path/to/project
# Read one Session chronologically, preserving queued human input
python3 scripts/read_claude_session.py --session <session-id> --project /path/to/project --full
# Export recent human wording grouped by Session
python3 scripts/extract_user_messages.py ./user-words --days 7 --group-by session
# Search sessions for keywords
python3 scripts/analyze_sessions.py search /path/to/project \
"ComponentName" "featureX" --from-date 2026-03-01 --to-date 2026-04-30
# Recover deleted files from the exact path printed by search
python3 scripts/recover_content.py <printed-session-path> -k DeletedComponent -o ./recovered/
# Search every project for multiple vanished job artifacts
python3 scripts/analyze_sessions.py search --all-projects \
artifact-a.html artifact-b.html \
--exclude-session <current-session-id>
# Get session statistics
python3 scripts/analyze_sessions.py stats /path/to/session.jsonl --show-files
🎬 Live Demo
Coming soon
📚 Documentation: See daymade-claude-code/read-claude-code-history/references/ for:
session_file_format.md - JSONL structure and extraction patternsworkflow_examples.md - Detailed recovery and analysis workflowsInstall:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:docs-cleaner)
Consolidate redundant documentation while preserving all valuable content.
When to use:
Key features:
🎬 Live Demo
Coming soon
Find Skills in configurable local repositories first, then approved sources and the CCPM registry. Preserve existing CCPM installation and management commands.
When to use:
Key features:
ccpm search <query>ccpm install <skill-name>@version syntaxccpm info <skill-name>Example usage:
# Search configured local repositories first (<skill-dir> is this loaded Skill)
uv run --script <skill-dir>/scripts/local_sources.py search pdf --tier owned
# Expand to the external registry when needed
ccpm search pdf # Find PDF-related skills
ccpm search "code review" # Find code review skills
# Install skills
ccpm install skill-creator # From registry
ccpm install daymade/skill-creator # From GitHub
ccpm install skill-creator@1.0.0 # Specific version
# List and manage
ccpm list # List installed skills
ccpm info skill-creator # Get skill details
ccpm uninstall pdf-processor # Remove a skill
# Install bundles
ccpm install-bundle web-dev # Install web development skills bundle
🎬 Live Demo
Coming soon
📚 Documentation: See daymade-skill/skills-search/SKILL.md for configuration, coverage and commands
Requirements: Python 3.10+, Git and uv for local search; CCPM CLI only for external registry/install/manage operations
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:pdf-creator)
Create professional PDF documents from markdown with proper Chinese typography using WeasyPrint.
When to use:
Key features:
Example usage:
uv run --with weasyprint scripts/md_to_pdf.py input.md output.pdf
🎬 Live Demo
Coming soon
📚 Documentation: See daymade-docs/pdf-creator/SKILL.md for setup and workflow details.
Requirements: Python 3.8+, pandoc (system install), weasyprint (or Chrome as fallback backend)
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:claude-md-progressive-disclosurer)
Optimize user CLAUDE.md files using progressive disclosure to reduce context bloat while preserving critical rules.
When to use:
Key features:
Example usage:
"Optimize my ~/.claude/CLAUDE.md using progressive disclosure and propose a plan."
🎬 Live Demo
Coming soon
📚 Documentation: See claude-md-progressive-disclosurer/SKILL.md.
Configure and run LLM evaluations with Promptfoo for prompt testing and model comparisons.
When to use:
Key features:
Example usage:
npx promptfoo@latest init
npx promptfoo@latest eval
npx promptfoo@latest view
🎬 Live Demo
Coming soon
📚 Documentation: See promptfoo-evaluation/references/promptfoo_api.md.
Requirements: Node.js (Promptfoo via npx promptfoo@latest)
Install:
claude plugin install daymade-macos@daymade-skills(suite-only — invoked asdaymade-macos:developing-ios-apps)
Build, configure, and debug iOS apps with XcodeGen, SwiftUI, and Swift Package Manager.
When to use:
project.ymlKey features:
Example usage:
xcodegen generate
xcodebuild -destination 'platform=iOS Simulator,name=iPhone 17' build
🎬 Live Demo
Coming soon
📚 Documentation: See developing-ios-apps/references/xcodegen-full.md.
Requirements: macOS + Xcode, XcodeGen
Fetch Twitter/X post and article content. Single-post text goes through the
fxtwitter mirror API (login-free, key-free, full long-form body); X Articles
with images go through fetch_article.py; the Jina API path remains as a
fallback with a known intermittent-ban risk.
When to use:
Key features:
tweet.text body including note_tweet long-formExample usage:
# Single post text (preferred): fxtwitter mirror, no key needed
curl -sS "https://api.fxtwitter.com/USER/status/TWEET_ID" \
| python3 -c "import json,sys; t=json.load(sys.stdin)['tweet']; print(t['text'])"
# X Article with images (full mode)
uv run --with pyyaml python scripts/fetch_article.py \
"https://x.com/USER/status/TWEET_ID" ./Clippings
# Jina fallback (intermittent — see SKILL.md risk note)
export JINA_API_KEY="your_api_key_here"
curl "https://r.jina.ai/https://x.com/USER/status/TWEET_ID" \
-H "Authorization: Bearer ${JINA_API_KEY}"
# Batch fetch multiple tweets
scripts/fetch_tweets.sh \
"https://x.com/user/status/123" \
"https://x.com/user/status/456"
# Fetch to file using Python script
python scripts/fetch_tweet.py https://x.com/user/status/123 output.md
🎬 Live Demo
Coming soon
📚 Documentation: See twitter-reader/SKILL.md for full details and URL format support.
Requirements:
Install:
claude plugin install daymade-macos@daymade-skills(suite-only — invoked asdaymade-macos:macos-cleaner)
The safest way to reclaim disk space on macOS. Start with targeted read-only diagnosis when a subsystem is already suspect—including Apple Content Caching—then analyze caches, application remnants, large files, and development environments only as needed.
Why macos-cleaner stands out:
Our design principles:
When to use:
Caching needs more space, an unlimited cache, or unexpectedly high ActualCacheUsedKey features:
CacheUsed from physical ActualCacheUsed, verifies peers and protected services, and uses Apple's supported limit/deactivate/flush controlsfdupes scans inside exact approved pathsWhat makes us different:
.git directories, node_modules, and build artifacts.Example usage:
# Install the Apple platform suite
claude plugin install daymade-macos@daymade-skills
# Ask Claude Code to analyze your Mac
"My Mac is running out of space, help me analyze what's using storage"
# Claude will:
# 1. Route a named suspect to targeted read-only diagnosis; scan broadly only if the source is unknown
# 2. Present categorized findings with safety levels
# 3. Explain each category (caches, remnants, large files, dev tools)
# 4. Recommend cleanup approach
# 5. Execute ONLY what you confirm
# Example analysis output:
📊 Disk Space Analysis
━━━━━━━━━━━━━━━━━━━━━━━━
Total: 500 GB
Used: 450 GB (90%)
Available: 50 GB (10%)
🟢 Safe to Clean (95 GB):
- System caches: 45 GB (apps regenerate automatically)
- Homebrew cache: 5 GB (reinstalls when needed)
- npm cache: 3 GB (safe to clear)
- Old logs: 8 GB (diagnostic data only)
- Trash: 34 GB (already marked for deletion)
🟡 Review Recommended (62 GB):
- Large downloads: 38 GB (may contain important files)
- App remnants: 8 GB (verify apps are truly uninstalled)
- Docker images: 12 GB (may be in use)
- Old .git repos: 4 GB (verify project is archived)
🔴 Keep Unless Certain (0 GB):
- No high-risk items detected
Recommendation: Start with 🟢 Safe items (95 GB), then review 🟡 items together.
🎬 Live Demo
Coming soon
📚 Documentation: See macos-cleaner/references/ for:
apple_content_caching.md - Targeted Apple Content Caching diagnosis and supported repaircleanup_targets.md - Detailed explanations of every cleanup targetmole_integration.md - How to combine scripts with Mole visual toolsafety_rules.md - Comprehensive safety guidelines and what to never deleteRequirements:
Verify factual claims in documents using web search and official sources, then propose corrections with user confirmation.
When to use:
Key features:
Example usage:
# Install the skill
claude plugin install fact-checker@daymade-skills
# Fact-check a document
"Please fact-check this section about AI model capabilities"
# Verify technical specs
"Check if these Claude model specifications are still accurate"
# Update outdated info
"Verify and update the version numbers in this documentation"
🎬 Live Demo
Coming soon
📚 Documentation: See fact-checker/SKILL.md for full workflow and claim types.
Requirements:
Review and improve Claude Code skills against official best practices with three powerful modes.
When to use:
Key features:
Example usage:
# Install the skill
claude plugin install daymade-skill@daymade-skills
# Self-review your skill
"Validate my skill at ~/my-skills/my-awesome-skill"
# Review external skill repository
"Review the skills at https://github.com/user/skill-repo"
# Auto-PR improvements
"Fork, improve, and submit PR for https://github.com/user/skill-repo"
🎬 Live Demo
Coming soon
📚 Documentation: See daymade-skill/skill-reviewer/references/ for:
evaluation_checklist.md - Complete skill evaluation criteriapr_template.md - Professional PR description templateStrategic guide for becoming an effective GitHub contributor and building your open-source reputation.
When to use:
Key features:
Contribution types explained:
Level 1: Documentation fixes (lowest barrier, high impact)
↓ (build familiarity)
Level 2: Code quality (medium effort, demonstrates skill)
↓ (understand codebase)
Level 3: Bug fixes (high impact, builds trust)
↓ (trusted contributor)
Level 4: Feature additions (highest visibility)
↓ (potential maintainer)
Example usage:
# Install the skill
claude plugin install github-contributor@daymade-skills
# Find good first issues
"Help me find projects with good first issues in Python"
# Write a high-quality PR
"Guide me through creating a PR for this bug fix"
# Build contribution strategy
"Help me plan a contribution strategy for building my GitHub profile"
🎬 Live Demo
Coming soon
📚 Documentation: See github-contributor/references/ for:
pr_checklist.md - Complete PR quality checklistproject_evaluation.md - How to evaluate projects for contributioncommunication_templates.md - Issue and PR communication templatesComplete internationalization/localization setup and auditing for UI codebases. Configure i18n frameworks, replace hard-coded strings with translation keys, ensure locale parity between en-US and zh-CN, and validate pluralization and formatting.
When to use:
Key features:
Example usage:
# Install the skill
claude plugin install i18n-expert@daymade-skills
# Setup i18n for a new project
"Set up i18n for my React app with English and Chinese support"
# Audit existing i18n implementation
"Audit the i18n setup and find missing translation keys"
# Replace hard-coded strings
"Replace all hard-coded strings in this component with i18n keys"
🎬 Live Demo
Coming soon
📚 Documentation: See i18n-expert/SKILL.md for complete workflow and architecture guidance.
Requirements:
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:claude-skills-troubleshooting)
Diagnose and resolve Claude Code plugin and skill configuration issues. Debug plugin installation, enablement, and activation problems with systematic workflows.
When to use:
Key features:
Example usage:
# Run diagnostic
python3 scripts/diagnose_plugins.py
# Batch enable missing plugins
python3 scripts/enable_all_plugins.py daymade-skills
🎬 Live Demo
Coming soon
📚 Documentation: See claude-skills-troubleshooting/SKILL.md for complete troubleshooting workflow and architecture guidance.
Requirements: None (uses Claude Code built-in Python)
Install:
claude plugin install daymade-audio@daymade-skills(suite-only — invoked asdaymade-audio:audio-router)
Selects the installed StepFun ASR, StepFun TTS, or transcript-to-minutes specialist and reads its full instructions at use time. Their original slash commands remain available manually. General audio transcription and transcript correction retain direct automatic entries.
Install:
claude plugin install daymade-audio@daymade-skills(suite-only — invoked asdaymade-audio:meeting-minutes-taker)
Transform meeting transcripts into high-fidelity, structured meeting minutes with iterative human review.
When to use:
Key features:
Example usage:
# Install the full audio suite (includes meeting-minutes-taker)
claude plugin install daymade-audio@daymade-skills
# Then provide a meeting transcript and request minutes
🎬 Live Demo
Coming soon
📚 Documentation: See daymade-audio/meeting-minutes-taker/SKILL.md for complete workflow and template guidance.
Requirements: None
Generate research reports from a durable study record: original sources, claim-level citations, rejected leads, and prior studies remain available for the next question. When several AI products or modes investigate one decision, coordinate their existing Skills and agents in parallel, preserve original outputs, and synthesize against the underlying sources and business question.
When to use:
Key features:
Example usage:
# Install the skill
claude plugin install deep-research@daymade-skills
# Then provide a report spec or template and request a deep research report
🎬 Live Demo
Coming soon
📚 Documentation: See deep-research/SKILL.md, research-asset-contract.md, and research_report_template.md.
Requirements: None
Discover, clone, update, and analyze competitor repositories with evidence-based competitive intelligence. Repository-backed findings must come from local cloned code; market-landscape claims must cite their source and volatility.
When to use:
Key features:
$HOME/workspace/competitors/{product}/file:line_number)Example usage:
# Install the skill
claude plugin install competitors-analysis@daymade-skills
# Then ask Claude to analyze a competitor
"分析竞品 https://github.com/org/repo"
"添加竞品到 flowzero 产品的竞品列表"
"看看 claude-flow-viewer 这个方向最近有哪些竞品更新"
🎬 Live Demo
Coming soon
📚 Documentation: See competitors-analysis/SKILL.md and competitors-analysis/references/ for templates.
Requirements: Git (for cloning repositories)
Diagnose and fix conflicts when using Tailscale alongside proxy/VPN tools (Shadowrocket, Clash, Surge) on macOS. Covers four independent conflict layers with specific guidance for SSH access to WSL instances.
When to use:
git push/pull fails with "failed to begin relaying via HTTP"operation not permittedKey features:
check vs accept)Example usage:
# Install the skill
claude plugin install tunnel-doctor@daymade-skills
# Then ask Claude to diagnose
"Tailscale ping works but SSH times out"
"Fix Tailscale and Shadowrocket route conflict on macOS"
"git push fails with failed to begin relaying via HTTP"
"Set up Tailscale SSH to my WSL instance"
🎬 Live Demo
Coming soon
📚 Documentation: See tunnel-doctor/references/proxy_conflict_reference.md for per-tool configuration and conflict architecture.
Diagnose Windows App (Microsoft Remote Desktop / Azure Virtual Desktop / W365) connection quality issues on macOS, with focus on transport protocol optimization (UDP Shortpath vs WebSocket fallback).
When to use:
Key features:
Example usage:
# Install the skill
claude plugin install windows-remote-desktop-connection-doctor@daymade-skills
# Then ask Claude to diagnose
"My VDI connection shows WebSocket instead of UDP, RTT is 165ms"
"Diagnose why RDP Shortpath is not working"
"Windows App transport protocol stuck on WebSocket"
🎬 Live Demo
Coming soon
📚 Documentation: See windows-remote-desktop-connection-doctor/references/ for log analysis patterns and AVD transport protocol details.
Run a scalable, evidence-driven product audit using parallel Claude Code agents and optional Codex CLI parallelization. Covers UX, API, architecture, and competitive benchmark workflows with quantified findings and priority recommendations.
When to use:
Key features:
codex availability)full, ux, api, arch, comparecompetitors-analysisExample usage:
# Install the skill
claude plugin install product-analysis@daymade-skills
# Then ask Claude for analysis
"Run product-analysis in full mode for launch audit"
"Do a UX audit and report quantified navigation findings"
"Run API audit and identify unused endpoints"
"Compare this product with our top competitors"
🎬 Live Demo
Coming soon
📚 Documentation: See product-analysis/SKILL.md and product-analysis/references/analysis_dimensions.md for dimension definitions and workflow guidance.
Requirements: Optional codex CLI (for multi-model parallel mode). Skill runs with Claude only if codex is not installed.
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:financial-data-collector)
Collect real financial data for any US publicly traded company from free public sources (yfinance). Output structured JSON with market data, historical financials (income statement, cash flow, balance sheet), WACC inputs, and analyst estimates - ready for downstream DCF modeling, comps analysis, or earnings review.
When to use:
Key features:
null with _source attribution, never default valuesExample usage:
# Install the suite
claude plugin install daymade-financial@daymade-skills
# Then ask Claude to collect data
"Collect financial data for META"
"Get financials for AAPL --years 3"
"Pull DCF inputs for NVDA"
🎬 Live Demo
Coming soon
📚 Documentation: See financial-data-collector/SKILL.md, output-schema.md, and yfinance-pitfalls.md.
Requirements: Python 3.11+, yfinance, pandas (auto-installed via uv inline dependencies).
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:excel-automation)
Create professionally formatted Excel files, parse complex .xlsm models with stdlib XML/ZIP workflows, and control Microsoft Excel windows on macOS via AppleScript.
When to use:
.xlsm files that fail in openpyxlKey features:
openpyxlzipfile + xml.etree (no heavy dependencies)definedNames repair workflow for problematic filesExample usage:
# Install the documentation suite
claude plugin install daymade-docs@daymade-skills
# Then ask Claude to automate Excel workflows
"Create a formatted valuation template workbook"
"Parse this .xlsm and extract the DCF sheet"
"Generate an AppleScript sequence to zoom and scroll Excel before screenshot"
🎬 Live Demo
Coming soon
📚 Documentation: See excel-automation/SKILL.md and formatting-reference.md.
Requirements: Python 3.8+, uv, openpyxl (auto via uv run --with openpyxl), macOS for AppleScript window control.
Install:
claude plugin install daymade-macos@daymade-skills(suite-only — invoked asdaymade-macos:capture-screen)
Capture application windows by CGWindowID with a reliable three-step workflow: discover window IDs via Swift, control app state via AppleScript, and capture outputs with screencapture.
When to use:
Key features:
CGWindowListCopyWindowInfo)Example usage:
# Install the Apple platform suite
claude plugin install daymade-macos@daymade-skills
# Then ask Claude to capture windows programmatically
"Find the Excel window ID and capture it silently"
"Create a multi-shot capture workflow for this workbook"
"Capture Chrome window sections with scripted scrolling"
🎬 Live Demo
Coming soon
📚 Documentation: See capture-screen/SKILL.md.
Install:
claude plugin install daymade-macos@daymade-skills(suite-only — invoked asdaymade-macos:macos-permissions)
Use for repeated TCC prompts, silent denials, and background jobs that cannot read protected files. It identifies the actual requester, checks for a usable existing grant, and verifies the repair through the real job. The Skill contains the diagnostic and Full Disk Access repair procedure.
📚 Documentation: See macos-permissions/SKILL.md.
Requirements: macOS. Reading TCC.db also requires Full Disk Access for the process doing the read.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:continue-claude-code-work)
Continue a verified Claude Code Session without reopening the old interactive
session. The continuation layer first consumes read-claude-code-history, then
reconstructs the original business outcome, unfulfilled requests, corrections,
proven prior assets, and one next action that directly advances the goal.
When to use:
.claude JSONL files instead of running claude --resumeKey features:
Example usage:
# Then ask Claude to resume from local artifacts
"continue work from session 123e4567-e89b-12d3-a456-426614174000"
"don't resume, just read the .claude files and continue"
"check what I was working on in the last session and keep going"
📚 Documentation: See continue-claude-code-work/SKILL.md.
Requirements: Python 3.8+, git for workspace reconciliation.
Install, troubleshoot, and use Scrapling CLI with a verified static-first workflow for extracting HTML, Markdown, or text from webpages. Includes a diagnostic script for broken extras installs, Playwright browser runtime checks, and smoke tests against real URLs.
When to use:
uv tool install scrapling, or scrapling extractmp.weixin.qq.com)Key features:
diagnose_scrapling.py script for CLI, browser runtime, and live URL smoke testsextract get, escalate to extract fetch only when needed#js_content for clean article Markdownclick, Playwright runtime setup, and curl: (60) trust-store failuresExample usage:
# Install the skill
claude plugin install scrapling-skill@daymade-skills
# Then ask Claude to work through Scrapling for you
"Install Scrapling CLI and verify the setup"
"Extract this WeChat article into Markdown with Scrapling"
"Decide whether this page needs static or browser-backed fetching"
🎬 Live Demo
Coming soon
📚 Documentation: See scrapling-skill/SKILL.md and scrapling-skill/references/troubleshooting.md.
Requirements: Python 3.6+, uv, Scrapling CLI, and Playwright browser runtime for browser-backed fetches.
One-stop wrapper for the official Tencent IMA skill (ima.qq.com). Installs upstream ima-skill to Claude Code, Codex, and OpenClaw via npx skills add, guides API key setup, detects and repairs known upstream issues under user consent, and implements a personalized fan-out search strategy that floats priority knowledge bases to the top.
When to use:
Skipped loading skill(s) due to invalid SKILL.md warnings related to ima-skillKey features:
~/.config/ima/{client_id, api_key} with env-var fallbackscripts/diagnose.sh read-only health check (install presence, credential liveness, known issues)scripts/search_fanout.py client-side cross-KB search with priority lists, subset-skip lists, and 100-hit silent-truncation detectionMODULE.md or prepend minimal frontmatter)~/.config/ima/copilot.json with illustrative-only template valuesExample usage:
# Install the skill
claude plugin install ima-copilot@daymade-skills
# Then ask Claude to drive the flow
"Install ima-skill and configure my IMA API key"
"Run diagnose on my ima-skill and fix whatever is broken"
"Search my IMA knowledge bases for embedding model comparisons, priority to my curated KB"
🎬 Live Demo
Coming soon
📚 Documentation: See ima-copilot/SKILL.md and ima-copilot/references/known_issues.md.
Requirements: Node.js 18+ (for npx skills), curl, unzip, Python 3.6+. IMA OpenAPI credentials from https://ima.qq.com/agent-interface.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:fixing-claude-export-conversations)
Reconstruct broken line wrapping in Claude Code exported .txt conversation files. Rebuilds tables, paragraphs, paths, and tool calls that were hard-wrapped at fixed column widths, and ships with an automated 53-check validation suite (file-agnostic, catches over- and under-merging regressions).
When to use:
YYYY-MM-DD-HHMMSS-*.txt/export output before sharing or archiving itKey features:
fix-claude-export.py) with --stats mode for before/after metricsvalidate-claude-export-fix.py) that catches regressionsuv and Python 3.8+Example usage:
# Fix and show stats
uv run daymade-claude-code/claude-export-txt-better/scripts/fix-claude-export.py broken.txt --stats
# Custom output path
uv run daymade-claude-code/claude-export-txt-better/scripts/fix-claude-export.py broken.txt -o fixed.txt
# Validate the fix
uv run daymade-claude-code/claude-export-txt-better/scripts/validate-claude-export-fix.py broken.txt fixed.txt
🎬 Live Demo
Coming soon
📚 Documentation: See claude-export-txt-better/SKILL.md and the bundled evals/ fixtures.
Requirements: Python 3.8+, uv package manager.
Export and sync Douban (豆瓣) book / movie / music / game collections to local CSV files via the reverse-engineered Frodo API. Full export covers all history; RSS incremental sync keeps daily updates current. No login, no cookies, no browser — just a user ID and it works.
When to use:
Key features:
.gitleaks.toml allowlist for the public Android APK credentialsExample usage:
# Full export of user's collections
uv run douban-skill/scripts/douban-frodo-export.py <douban-user-id>
# Incremental RSS sync (last ~10 items per category)
uv run douban-skill/scripts/douban-rss-sync.py <douban-user-id>
🎬 Live Demo
Coming soon
📚 Documentation: See douban-skill/SKILL.md and douban-skill/references/troubleshooting.md for the complete failure log of rejected approaches.
Requirements: Python 3.8+, uv package manager. No login or cookies required.
Designs and diagnoses safe Terraform releases as well as the provisioner traps learned from real incidents. It keeps staging and production on one required configuration schema, validates exact candidate bytes + the Compose-rendered environment + the immutable runtime before live mutation, and binds saved plans to staging evidence, source provenance, explicit production authorization, and independent readback.
When to use:
null_resource provisioners or remote-exec blocks that SSH into fresh instancesterraform applyKey features:
Example usage:
# Trigger the skill naturally during Terraform work
"I'm getting 'docker: not found' in my null_resource provisioner after apply"
"My rsync local-exec is failing with 'connection unexpectedly closed'"
"Help me write a multi-env Terraform setup without snapshot cross-contamination"
"Staging has this Caddy variable but production leaves it empty — how do I validate both safely?"
🎬 Live Demo
Coming soon
📚 Documentation: See terraform-skill/SKILL.md and the bundled references/ for detailed remediation patterns.
Requirements: None (Terraform-adjacent knowledge only; no runtime dependencies).
Guides users through structured narrative design (ABCDEFG model), then delegates visual generation to baoyu-slide-deck. Focuses on what machines can't do — narrative co-design with humans.
When to use:
Key features:
Example usage:
# Trigger the skill naturally
"Help me turn my article into a slide deck"
"Create a presentation from my talk transcript"
"I need a 20-minute deck for a workshop"
Requirements: baoyu-slide-deck skill for visual generation.
Batch-place existing images onto an Excalidraw whiteboard, laid out on a generous grid so nobody has to drag them apart afterwards. Also turns a slide deck into clean per-slide images first, and inspects what is inside a .excalidraw file. The available Excalidraw MCP servers and skills cover element CRUD and export but document no image element type, no dataURL handling, and no files map — embedding your own pictures is the gap this fills.
When to use:
Key features:
--exclude for images already on the board--template-from copies the image-element field set out of your own board — Excalidraw's published schema stops before fileId/status/scale/cropinnerText still reads hidden fragments as presentExample usage:
# Trigger the skill naturally
"Put these screenshots on my Excalidraw board"
"Add my old workshop images to the canvas, spaced out"
"Turn this deck into images I can draw on"
Note: Not for generating a diagram from a text description — that is a different job.
Falsification-first methodology for network, streaming, and protocol-layer bugs where the obvious cause is probably wrong. Built from a real 5-hour SSE incident where assumption-stacking wasted hours that a 10-minute layered experiment would have resolved.
When to use:
ECONNRESET, HTTP/2 RST_STREAM, INTERNAL_ERROR)ERR_CONNECTION_CLOSED, SSL_ERROR_SYSCALL, fake TUN DNS IPs, CNAME-based rule overrides)UNKNOWN_CERTIFICATE_VERIFICATION_ERROR, wrong-site certificate)Key features:
layered-isolation-probe.sh, mock-idle-upstream.py)ERR_CONNECTION_CLOSEDRequirements: None (methodology + portable shell/Python probes).
Install:
claude plugin install daymade-audio@daymade-skills(suite-only — invoked asdaymade-audio:stepfun-tts)
Generate Chinese / Japanese speech with stepaudio-2.5-tts. Captures the two non-obvious TTS pitfalls that cost hours otherwise: voice_label removal (replaced by natural-language instruction) and stricter 2.5-era censorship (死/消失/political terms).
When to use:
censorship_block fallbackstep-tts-2 to stepaudio-2.5-tts (voice_label → instruction breaking change)Key features:
stepaudio-2.5-tts with instruction (≤200 chars natural-language mood) + inline () prosodytts_generate.py (with --batch <jsonl>) and ab_compare.sh$STEPFUN_API_KEY → ${CLAUDE_PLUGIN_DATA}/config.json fallbackreferences/migration_from_v2.mdRequirements: StepFun API key, "Normal" tier (https://platform.stepfun.com/). For ASR / transcription, use the sibling stepfun-asr skill below.
Install:
claude plugin install daymade-audio@daymade-skills(suite-only — invoked asdaymade-audio:stepfun-asr)
Transcribe Chinese / English audio with stepaudio-2.5-asr. Hides the #1 trap of the 2.5 ASR family: it does NOT live on /v1/audio/transcriptions — the wrong endpoint returns a misleading model stepaudio-2.5-asr not supported error that looks identical to a permission/whitelist failure.
When to use:
step-asr / step-asr-1.1 (different endpoint, different body shape, SSE response)model stepaudio-2.5-asr not supported error (= wrong endpoint, not permission)Key features:
/v1/audio/asr/sse SSE streaming with base64 audio + nested JSON body (the script handles all four traps)asr_transcribe.py — pure-stdlib CLI, auto-detects mp3/wav/ogg/opus/pcm by extensionerror events (censorship can fire on ASR side too — rare but real)$STEPFUN_API_KEY → ${CLAUDE_PLUGIN_DATA}/config.json fallbacktranscript-fixer (ASR error correction) and meeting-minutes-taker (structured minutes) as natural downstream skillsRequirements: StepFun API key, "Normal" tier (https://platform.stepfun.com/). Plan keys cannot call audio endpoints.
Make a repository runnable and handoff-ready without guessing its stack or changing how collaborators normally work. The skill reads project authority first, repairs the verified gap, and treats startup instructions, lifecycle hooks, and Git mutation as different mechanisms with different safety boundaries.
When to use:
Key features:
check_env.py infers only declared toolchains from manifests/lockfiles; it does not assume ffmpeg, uv, Python, or .envExample usage:
# Install the skill
claude plugin install auto-repo-setup@daymade-skills
# Then ask Claude naturally
"我跑不起来这个仓库"
"帮我设置一下这个项目的环境"
"进入项目先同步远端；本地有改动不要自动 stash"
"为什么同一个 SessionStart 输出了三次？先查清来源"
"只有首条消息前必须注入动态提醒时，才帮我装 hook"
"git push 被拒了"
Requirements: The guidance itself has no runtime dependency. Bundled Python utilities require Python 3.10+; no external API key is required.
Render a terminal CLI program's colored output to a PNG so Claude can actually see the rendered result — color contrast, alignment, background blocks, highlighting — instead of only reading plain text and raw ANSI escape codes. Reading a hex value is guessing; seeing the rendered contrast on the real terminal background is verification.
When to use:
Key features:
render_ansi.sh (freeze/Chrome auto-select), ansi2html.py (stdlib renderer)Example usage:
# terminal-screenshot lives in the daymade-claude-code suite
claude plugin install daymade-claude-code@daymade-skills
# Then ask Claude naturally
"verify my delta diff colors"
"看一下这个终端配色的真实效果"
"is the add/remove contrast in git diff strong enough?"
Requirements: macOS. charmbracelet/freeze (preferred renderer) or Google Chrome (fallback). Python 3 for the fallback renderer.
Convert a PDF into one self-contained, readable HTML file that preserves images, charts and reading order — optionally translating it into another language while keeping every figure. A PDF is a layout, not just a text stream, so the workflow renders each page for you to see before building, and renders the HTML for visual verification before delivery.
When to use:
Key features:
Example usage:
# pdf-to-html lives in the daymade-docs suite
claude plugin install daymade-docs@daymade-skills
# Then ask Claude naturally
"把这个 PDF 转成中文网页版"
"make this report readable as HTML"
"translate this PDF to English but keep the charts"
Requirements: uv, Google Chrome or Chromium (visual verification). Python packages (PyMuPDF, Pillow, numpy) auto-install via uv run --with.
Install:
claude plugin install daymade-audio@daymade-skills(suite-only — invoked asdaymade-audio:asr-transcribe-to-text)
Transcribe audio and video files to text using Qwen3-ASR via two interchangeable inference paths: local MLX on macOS Apple Silicon (no API key, 15-27x realtime) or a remote vLLM/OpenAI-compatible API for any platform. Auto-detects the platform and recommends the best path, persisting the choice in ${CLAUDE_PLUGIN_DATA}/config.json.
When to use:
Key features:
transcribe_local_mlx.py loads the model once and processes files sequentially (no GPU contention)overlap_merge_transcribe.py splits into 18-minute chunks with 2-minute overlap and fuzzy-mergestranscript-fixer to clean ASR recognition errors on the outputExample usage:
# asr-transcribe-to-text lives in the daymade-audio suite
claude plugin install daymade-audio@daymade-skills
# Then ask Claude naturally
"transcribe this meeting recording to text"
"把这个录音转成文字"
"convert lecture.mp4 to a transcript"
Requirements: uv, ffmpeg/ffprobe. Local MLX path needs macOS Apple Silicon; remote path needs a reachable vLLM/OpenAI-compatible ASR endpoint. No API key for local mode.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:marketplace-dev)
Create and maintain Claude Code plugin marketplaces: convert a repository, consolidate standalone skills into a new or existing suite, move skills between suites, validate real installation/cache boundaries, and ship the result through a PR.
When to use:
claude plugin installmarketplace.json (plugin distribution, one-click install, auto-update)Unrecognized key: "$schema" or duplicate plugin namesKey features:
$schema is rejected, metadata has only 3 valid fields, strict: false semantics, single-skill vs suite source/skills patternscheck_marketplace.sh runs four checks (JSON syntax → claude plugin validate → source/skills resolution → reverse sync) and exits non-zero on failuresource produced the intended snapshotmarketplace.json edit; warn on un-bumped version when a SKILL.md changes) that auto-activate with the pluginExample usage:
# marketplace-dev lives in the daymade-claude-code suite
claude plugin install daymade-claude-code@daymade-skills
# Then ask Claude naturally
"turn this skills repo into a plugin marketplace"
"generate a marketplace.json for this repo and validate it"
"add my new skill to the marketplace and open a PR"
"move these standalone skills into daymade-macos and make them suite-only"
Requirements: claude CLI (for claude plugin validate / install tests), jq. Git remotes configured if opening an upstream PR.
Install:
claude plugin install daymade-skill@daymade-skills(suite-only — invoked asdaymade-skill:skill-creator)
The essential meta-skill for building your own skills. It scales verification to the change: bounded fixes get targeted checks, narrow behavior changes get sampled replays, and broad/high-risk work becomes eligible for heavier evidence without automatically starting it. Paired baselines, graders, benchmarks, and viewers require explicit authorization or a decision-bearing evidence plan plus opt-in. It also supports explicit benchmarking and optimizes a skill's description for better triggering accuracy.
When to use:
Key features:
context: fork decision guide (subagents can't spawn subagents or call skills) and composable/orthogonal skill designinit_skill.py scaffolding, package_skill.py (auto-validates), and security_scan.py (gitleaks-based secret/PII detection)Example usage:
# skill-creator lives in the daymade-skill suite
claude plugin install daymade-skill@daymade-skills
# Then ask Claude naturally
"create a skill that does X"
"improve this skill's description so it triggers more reliably"
"benchmark this skill against a no-skill baseline"
Requirements: Python 3, uv, PyYAML (validation/packaging), gitleaks (security scan). claude CLI only for agent evals and description-optimization runs.
Extract Feishu (Lark) Docs, Wiki pages/collections, spreadsheets (including cell-attachment file download), and Minutes (妙记) transcripts into faithful local Markdown. The primary path uses the lark-cli API — it extracts the document body programmatically (no model paraphrasing), recursively follows a collection's reference graph, and reads permission boundaries from error codes; a browser-DOM path is the fallback only when lark-cli cannot reach the content.
When to use:
.docx into faithful Markdown with heading/highlight restorationKey features:
jq (never retyped by the model — the single most important fidelity rule)feishu_extract_refs.py, plus a residual rich-media-tag acceptance gate so no referenced doc is silently missed.docx → Markdown with font-size→heading and w:shd→highlight restoration, then visual verificationLARK_CLI_NO_PROXY=1 discipline for *.feishu.cn (avoids credential leak/DNS hijack) and a U+FFFD encoding-corruption final checkExample usage:
# Install the skill
claude plugin install feishu-doc-scraper@daymade-skills
# Then ask Claude naturally
"把这个飞书合集导出成 markdown"
"export this Feishu Minutes transcript"
"save this Lark wiki page as Markdown"
Requirements: lark-cli binary (npm @larksuite/cli) authenticated to the target tenant; jq. Fallback path needs a browser-automation surface; the docx path needs python-docx and a docx→md converter (the bundled doc-to-markdown skill or pandoc).
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:financial-router)
Routes Bigdata/RavenPack, free US company fundamentals, A-share news, sector Top N and announcements, and Gangtise setup to one bundled specialist. Pharmaceutical reporting and general argument review keep their direct Skills; all specialist slash commands remain available for manual use.
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:bigdata-skill)
Pull Bigdata.com (RavenPack) financial and news data through the official bigdata-client SDK and its public /v1/* REST endpoints — reaching the structured substrate the Bigdata MCP server doesn't hand over. The MCP returns prose chunks and pre-synthesized tearsheets; this toolkit reaches structured financials, prices, analyst estimates, a daily entity-sentiment series, annotated chunk search with sentiment + entity spans, and a screener.
When to use:
bd_v2_ API key, rp_entity_id, query_unit/chunk cost, bigdata-client, or "the bigdata MCP isn't enough"Key features:
BigdataClient exposing both the SDK (search + knowledge graph) and a REST escape hatch (bd._api.http) for every /v1/* endpoint the SDK never wrappedfields_values_to_records() to flatten {fields, values} responses1 query_unit = 10 chunks, only chunk-search billed, ChunkLimit (never a bare int), rerank thresholds, 50%-cheaper batch search, and a CostModel/CostTracker budget vetorc() SSL-retry wrapper for the common first-handshake SSL: UNEXPECTED_EOF, plus a known-pitfalls reference with reproductions and fixesBIGDATA_API_KEY (no plaintext fallback); read-only, never writes/uploadsExample usage:
# Install the suite
claude plugin install daymade-financial@daymade-skills
export BIGDATA_API_KEY=bd_v2_xxxxxxxx
# Then ask Claude naturally
"pull NVIDIA's forward analyst estimates and last earnings surprise from Bigdata"
"give me a daily entity-sentiment series for this ticker"
"the bigdata MCP only gave me a tearsheet — I need the structured fields"
Requirements: A bd_v2_ Bigdata.com API key (env var, never hardcoded), uv, the official bigdata-client SDK in an isolated venv. Optional outbound/WSS proxy only if your network needs one to reach api.bigdata.com.
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:gangtise-copilot)
One-command installer, credential configurator, and diagnostic layer for the full Gangtise (岗底斯投研) OpenAPI skill suite. Installs the official Gangtise skills (data, research, utility), configures accessKey/secretAccessKey with a live auth check, and runs a read-only health diagnostic — solving the suite's core discoverability problem (no public manifest, listing-disabled OBS bucket, parallel naming lines).
When to use:
gangtise-* skilltoken is invalid / 接口地址错误, or "my gangtise install is broken"Key features:
install_gangtise.sh downloads the declared OBS bundles → extracts skill directories → symlinks them into detected agent skills dirs (Claude Code, OpenClaw, Codex), with minimal/workshop/full/--only presetsconfigure_auth.sh writes a shared XDG credential file (mode 600), runs a live auth call, and symlinks every skill's .authorization to itdiagnose.sh reports install state, credential validity, and scoped capability tiers (auth scope vs RAG scope)Example usage:
# Install the suite
claude plugin install daymade-financial@daymade-skills
# Then ask Claude naturally
"装一下 gangtise 的所有 skill 并配置好凭据"
"my gangtise skills report token is invalid — diagnose it"
"宁德时代的研报用哪个 gangtise skill 查"
Requirements: A Gangtise accessKey + secretAccessKey; bash, curl, network access to the official OBS bucket and open.gangtise.com. Works with Claude Code, OpenClaw, and Codex agent layouts.
Co-create a personal investment-research LLM Wiki (Andrej Karpathy's pattern) where the user's OWN analysis framework becomes a living CLAUDE.md — built by interviewing them rather than handing over a template. Pure markdown + [[wikilinks]]; knowledge is compiled into pages instead of being re-retrieved from raw files on every query (Karpathy's core idea — no RAG pipeline in place of compilation; a local search engine such as qmd over the grown wiki is part of his pattern). The value is extracting the user's personal investment preferences into THEIR OWN schema, never imposing a standard one.
When to use:
Key features:
init_vault.py scaffolds the mechanism layer only (no schema), then an 8-dimension interview builds the user's own CLAUDE.md in their own wordslint-vault.py keep the vault consistent and fight derived-value driftanalyst-track-record skill and Bash) and chains into analyst-track-record for analyst back-testing — without rebuilding itExample usage:
# Install the skill
claude plugin install llm-wiki-setup@daymade-skills
# Then ask Claude naturally
"帮我搭一个投研第二大脑"
"build me a personal investment-research wiki in Karpathy's style"
"ingest this earnings call into my research vault"
Requirements: Python 3, uv (for init_vault.py / lint), git. Markdown + wikilinks; no vector DB or embedding service required (qmd is an optional local search layer once the wiki grows). Pairs with the analyst-track-record skill for back-testing.
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:benchmark-due-diligence)
Run adversarial due-diligence on a benchmark the user envies — a founder, KOL, company, or product whose claimed success looks inflated — separating marketing bubble from real signal, then mapping the validated playbook onto the user's own resources. The adversarial, decision-oriented cousin of deep-research: it assumes the picture is inflated until proven otherwise and ends in "what this means for ME", not a neutral report.
When to use:
deep-research's neutral briefingKey features:
坐实/存疑/证伪-水分 verdicts) → due-diligence conclusion (bubble-busting table + attribution breakdown) → commissioner resource-mappingdeep-research fan-out, osint-investigate identity checks, the qcc family for 工商 data, agent-reach for social-platform data)context: fork would silently break the fan-out)Example usage:
# Install the financial research suite
claude plugin install daymade-financial@daymade-skills
# Then ask Claude naturally
"帮我尽调一下这个创始人，他到底有没有水分"
"tear down this competitor's playbook and tell me what I can actually copy"
"this KOL claims 0-to-1M users — is that real, and is it replicable for me?"
Requirements: Web access for the collection/verification agents. Optionally composes with deep-research, osint-investigate, the qcc skill family, and agent-reach; renders a shareable report via pdf-creator.
Fetch real, citable data for any Bilibili (B站) video — title, UP follower count, publish date, tags, partition, per-part cids, live stats (view/like/coin/favorite/share/reply/danmaku), and full danmaku (bullet-comment) text — in one view/detail call, login-free. Built so engagement numbers are cheap to fetch and impossible to fake, instead of hand-typed into a doc where they rot.
When to use:
av number, b23.tv short link, or full URL — all normalized automaticallyKey features:
bili-fetch.sh returns full metadata + live stats + UP fans + tags + every part's cid; metrics carry a fetched_at timestamp because they drift in real timebili-danmaku.sh pulls and decompresses the danmaku full text; bili-subs.sh handles the login-gated subtitle track (asks before touching browser cookies)bili-selftest.sh health-check verifies every endpoint against the live API, so API drift surfaces as one clear FAIL instead of a silent wrong answerx/v3/fav/*, verified no-WBI), the verified SESSDATA subtitle path, and the WBI request-signing algorithm for space/wbi/* extensionExample usage:
# Install the skill
claude plugin install bilibili-source@daymade-skills
# Then ask Claude naturally
"pull the real view/like/favorite counts for this B站 video so I can cite them"
"这个 B站 视频弹幕里大家在说什么？"
"grab the subtitle transcript from this bilibili video so I can summarize it"
Requirements: curl, jq, python3 (danmaku decompression). yt-dlp only for the login-gated subtitle path. No login for stats / metadata / danmaku.
Preserve and recover recorded Ghostty CLI sessions through the canonical Skill. For scheduled metadata backups, partial coverage and operating prerequisites, use the automatic workflow.
claude plugin install ghostty-use@daymade-skills
# Then ask Claude naturally
"我要重启电脑更新，把终端会话都保存一下"
"恢复之前的 ghostty 窗口和会话"
"restore my ghostty tabs after the update"
"检查一下有没有会话丢了"
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:claude-usage-analyst)
Turn local ccusage data into an evidence-based, human-readable explanation of where your Claude Code / Claude Desktop tokens, cost, and quota went — separating observed numbers from interpretation instead of guessing.
When to use:
fable / opus / sonnet) is unusually expensive for your workloadKey features:
analyze_claude_usage.py summarizes tokens, cost, input/output, and cache create/read over any date window and timezone--model-a / --model-b) weighs both token volume and estimated cost — a model can be cheap per token but expensive overallccusage output; cache-read pressure is counted even when you never typed those tokensccusage claude measures local Claude Code logs, not a full Claude.ai chat billExample usage:
# Install the suite
claude plugin install daymade-claude-code@daymade-skills
# Then ask Claude naturally
"why did my Claude quota run out today?"
"is opus more expensive than sonnet for what I'm doing?"
"break down my Claude Code token usage for this month"
Requirements: ccusage (via npm i -g ccusage or npx ccusage@latest), python3.
claude plugin install marketplace-health-check@daymade-skills
Run a comprehensive, evidence-based health check of this skills marketplace repo with a parallel fan-out Dynamic Workflow — six inspectors cover code/script safety, documentation/SSOT consistency, security/PII leaks, open-PR triage, open-issue triage, and marketplace-manifest integrity at once — then the serious findings are Counter-Reviewed before they reach the report.
When to use:
Key features:
Example usage:
# Install
claude plugin install marketplace-health-check@daymade-skills
# Then ask Claude naturally
"do a full health check of this repo before I cut a release"
"audit the marketplace — code, docs, PII, PRs, issues, everything"
"全面体检一下这个仓库"
Requirements: gh CLI (authenticated), git, jq, python3; opt-in to the Workflow tool (asking to run the health check is the opt-in).
claude plugin install daymade-claude-code@daymade-skills
Set up multiple isolated Claude Code CLI profiles so you can run different LLM providers (Kimi, MiniMax, GLM, DeepSeek, StepFun, Anthropic) in separate terminal windows at the same time — each profile gets its own .claude.json state while sharing skills, projects, hooks, and agents.
When to use:
Key features:
~/.config/claude-switch-models-setup/ and seeds an empty Codex activation manifest only when none exists~/.claude/settings/<provider>.json templates with required isolation flagsclaude-profiles-init creates isolated ~/.claude-profiles/<provider>/ directories with symlinked shared resources~/.agents/skills; a declared active subset may keep legacy ~/.codex/skills compatibility linksExample usage:
# Install the suite
claude plugin install daymade-claude-code@daymade-skills
# Then ask Claude naturally
"set up Claude Code profiles for Kimi and DeepSeek"
"I want to run Kimi and Anthropic in separate terminals"
"install the multi-provider profile setup from the workshop"
Requirements: claude CLI, zsh or bash, python3, plus API keys for the providers you want to use.
claude plugin install llm-eval-harness@daymade-skills
Evaluate any LLM behind an OpenAI- or Anthropic-compatible endpoint across four dimensions — instead of trusting a vendor's headline numbers — and report measured results separately from inferred ones.
When to use:
Key features:
speed_probe.py — TTFT + thinking-aware tokens/sec), concurrency/stability (concurrency_probe.py — success rate, p50/p90, breaking point), Anthropic protocol compliance (protocol_probe.py — thinking-block trigger rate over N≥10), and quality regression (usecase_runner.py + independent blind judges)reasoning_content separately so reasoning tokens don't inflate tok/s (the trap that once read a ~750 tok/s model as 4700)fully-implemented / intermittent (k/N) / not-implemented, never concluding from a single sample, with Connection: close so a load balancer can't pin all samples to one replica--key-env MY_KEY) — the key never appears in ps, shell history, or a saved report~/.llm-eval/usecases.json) so it survives skill updates and never lands in a public repoExample usage:
export MY_KEY=sk-... # the key never appears in a command below
# Speed: real-task throughput + sustained decode ceiling
uv run --with openai python scripts/speed_probe.py \
--base-url https://api.example.com/v1 --model some-model --key-env MY_KEY --mode both
# Concurrency: ramp until it breaks
uv run --with aiohttp python scripts/concurrency_probe.py \
--url https://api.example.com/v1/chat/completions --model some-model \
--key-env MY_KEY --format openai --concurrency 10 20 40 60
🎬 Live Demo
Coming soon
📚 Documentation: See llm-eval-harness/references/evaluation_disciplines.md for the reasoning behind each discipline and llm-eval-harness/references/quality_blind_judge.md for the blind-judge method.
Requirements: Python 3.8+, uv; openai and aiohttp (auto-installed via uv run --with); an API key for the endpoint under test. Optionally composes with promptfoo-evaluation for rubric-based gating.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:read-claude-web-conversation)
Extract the complete active Claude.ai conversation through the Claude Code operations suite: message text, real tool calls and outputs, uploads, generated images, and sandbox deliverables. Render faithful Markdown locally, reconstruct files created through tool calls, or download original bytes through the logged-in browser session.
When to use:
Requirements: Installed daymade-claude-code suite and a logged-in Chrome
session. The AppleScript fallback requires macOS and one user-enabled Chrome
developer-menu toggle.
claude plugin install setup-notifications-via-wecom@daymade-skills
Set up reusable WeCom (Enterprise WeChat) webhook notifications for technical status reports, alerts, and completion messages. The target is explicitly classified: the user's own channel may send automatically; every other target requires human confirmation.
For recurring reports on an existing SLS pipeline, follow the bundled adapter guide for truncation limits, evaluation dates, recipient conflicts and native delivery acceptance.
When to use:
Requirements: WeCom bot webhook URL and shell access.
claude plugin install notify-wecom@daymade-skills
Send a single WeCom group-bot message using the configured target identity: self sends directly, while others requires human confirmation.
When to use:
/notify-wecomRequirements: WeCom bot webhook URL plus the explicit recipient scope/label and canonical sender binding created by the setup command.
claude plugin install github-sensitive-data-cleanup@daymade-skills
Scan and remove sensitive data from GitHub repository history, with backup, visibility checks, and force-push safety gates.
When to use:
Requirements: git, GitHub access, and the relevant scanning/history-rewrite tools for the target repository.
Install:
claude plugin install daymade-codex@daymade-skills(suite-only — invoked asdaymade-codex:codex-image-gallery)
Start a self-contained local web gallery for Codex-generated image outputs. The skill bundles its Node server and HTML UI, scans ~/.codex/generated_images by default, and can point at another folder with GALLERY_ROOT.
When to use:
~/.codex/generated_imagesKey features:
scripts/server.mjs and assets/index.html/api/images scan; no hardcoded manifest/images/<relative-path> image route with path traversal protectionGALLERY_ROOT, PORT, and HOSTRequirements: Node.js 18+ and access to the image folder.
claude plugin install frontend-visual-qa@daymade-skills
Audit the UI users can actually see, with explicit evidence levels and no source mutation by default.
When to use:
ui-designer/design work and the broader process managed by qa-expertKey features:
claude plugin install openclaw@daymade-skills
Manage OpenClaw (龙虾/lobster) instance configurations — audit, diff, copy, add-model, list, and switch models across openclaw.json files.
When to use:
Key features:
--no-audit escape hatchclaude plugin install download-gemini-images@daymade-skills
Download images (uploaded files or generated previews) from a Google Gemini conversation page using your logged-in Chrome session, then package them into an ordered ZIP.
When to use:
Key features:
pageAssets fallback when lightbox automation failsclaude plugin install wps-doc-scraper@daymade-skills
Faithfully archive public WPS / KDocs / 金山文档 links — especially embedded ProcessOn mind maps and canvases — as raw source data, original SVG/PNG, and Markdown, without logging in.
When to use:
kdocs.cn or wps.processon.com link to scrape, save, download, or convert to MarkdownKey features:
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:ashare-news-fetcher)
Aggregate A-share (Chinese stock market) news, policy, and sentiment from public sources — 财联社, 华尔街见闻, 金十, 新浪 7x24, 东财快讯, regulator announcements (CSRC / PBoC / SSE), 东方财富股吧 — into structured JSON or Markdown.
When to use:
Key features:
Install:
claude plugin install daymade-codex@daymade-skills(suite-only — invoked asdaymade-codex:local-codex)
Delegate coding tasks to the local OpenAI Codex CLI agent using your ChatGPT Pro OAuth flat-rate subscription. Wraps codex exec / codex review for code generation, refactoring, and review without per-token API charges.
When to use:
codex exec for non-interactive coding tasksKey features:
~/.codex/auth.json (no API keys)claude plugin install openclaw-model-switch@daymade-skills
Switch the default AI model for an OpenClaw instance (e.g., Kimi K2.6 → K2.7) by safely editing openclaw.json with automatic backup, model validation, and optional gateway restart.
When to use:
Key features:
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:pharma-daily-report)
Generate an A-share pharmaceutical sector daily report — pull real-time quotes for core pharma stocks from Sina Finance, rank 7 sub-sectors, top gainers/losers, and estimated fund flow, then optionally push a rich-text report via Feishu.
When to use:
Key features:
Install:
claude plugin install gemini-history-analyzer@daymade-skills
Analyze Google Takeout exports of Gemini conversation history — extract and categorize transcripts and attachments, mine them for domain-specific insights with context-verified keyword search, and optionally distill findings into a personal knowledge base.
When to use:
Key features:
unar, not the corrupting macOS unzip)Install:
claude plugin install daymade-skill@daymade-skills(suite-only — invoked asdaymade-skill:skill-governance)
Govern the Skill surface Claude Code and Codex actually expose without losing cold capability. It separates canonical source, installed inventory, discovery policy, the fresh model-visible catalog, and router-resolved runtime resources.
When to use:
~/.agents/skills without treating every disk bundle as activeKey features:
skills/list, detecting truncation, omitted enabled entries, identity/source collisions, disabled-path drift, scan errors, and missing direct activation entriesInstall:
claude plugin install daymade-docs@daymade-skills(suite-only; invoke asdaymade-docs:photo-to-scanned-pdf)
Turn phone photos of contracts, receipts, forms, certificates, or handwritten pages into a clean A4 PDF. The workflow combines perspective correction, scanner-style background cleanup, colored-paper handling, explicit content-based page ordering, and a mandatory whole-document contact-sheet check.
claude plugin install github-review-pr@daymade-skills
Review or re-review one named contributor PR—including a closed PR under reconsideration—or sweep every open PR newest-to-oldest as a repository maintainer, always against the live base rather than a stale PR snapshot.
When to use:
Key features:
PR, BASE, or SHARED before assigning repair ownershipDECLINE from a fix request or true supersessionExample usage:
/github-review-pr --personal-maintainer https://github.com/<owner>/<repo>/pull/<number>
/github-review-pr --personal-maintainer --all-open
main changed since the last review; tell me what would actually land now
review all of our open PRs newest to oldest and apply my prior maintainer principles
can we merge this contribution and fix the remaining repo bookkeeping ourselves?
Requirements: authenticated gh CLI, git with merge-tree --write-tree, and jq.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:read-codex-history)
Read, search, and export local Codex history without continuing the old task. It keeps the prompt ledger, state database, and rollout JSONL as three distinct evidence surfaces so exact user input, inventory metadata, and Agent behavior are never silently substituted for one another.
Key features:
session_meta.id equality and reports fused/missing rollout gaps instead of guessingExample usage:
/daymade-claude-code:read-codex-history
list the recent Codex chats for this folder
show my 200 most recent original Codex inputs, grouped by Session
read Codex Session 01abc... chronologically and show its fork lineage
show which Codex sessions in this workspace have held canonical writer-lock files
show Codex threads including archived conversations as JSON
📚 Documentation: See storage_and_portability.md for source selection, path normalization, privacy boundaries, and diagnostics.
Requirements: Python 3.10+; no third-party packages or network access.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:continue-codex-work)
Continue a prior Codex task without replaying it through codex resume.
The Skill first requires read-codex-history to prove selected identity, fork
lineage, chronology, compacted context, and gaps; it then rebuilds the original
business outcome, unfulfilled requests, user corrections, proven assets, and one
direct next action before changing the project.
It is for a new/different Agent context taking over an earlier rollout. When
Codex itself natively resumes the same conversation and prior context is already
present, continue directly instead of invoking this Skill.
/daymade-claude-code:continue-codex-work 019f66...
Codex was cut off mid-task; recover that rollout and finish the work
📚 Documentation: See continue-codex-work/SKILL.md.
Requirements: Python 3.10+ and local Codex rollout files.
Install:
claude plugin install git-safety-net@daymade-skills
Prevent and recover from the branch/stash/rebase tangles that strand or lose local commits, detached or dirty worktrees, and answer "is everything actually merged?" with content-level proof instead of misleading commit counts. Every command is non-destructive or additive until a step is explicitly labeled destructive.
Key features:
git reflog and git fsck (the ~90-day window)refs/dangling-backup/ before any cleanupExample usage:
/git-safety-net
did I lose any commits after all that branch switching?
is everything merged into main, or is something still stranded?
recover the commit I lost after a bad rebase
prove this worktree can be deleted without losing anything
📚 Documentation: See recovery_playbook.md, merge_verification.md, and prevention_practices.md.
Requirements: Git and a standard Bash shell. No third-party packages; current remote status requires network access, while offline audits use cached remote-tracking refs.
Install:
claude plugin install daymade-codex@daymade-skills(suite-only — invoked asdaymade-codex:design-style-picker)
Turn vague taste into concrete visual choices. Instead of guessing one final design, batch-generate a structured set of visual directions so the user picks the style they actually want — the goal is exposing the taste boundary quickly, not reading minds.
Key features:
Example usage:
/daymade-codex:design-style-picker
I can't describe the style I want — show me a batch of options
this draft is too colorful and too generic, generate more directions
evolve our current design system without throwing away existing assets
📚 Documentation: See selection-playbook.md.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:claude-migrate-memory-to-doc)
Migrate Claude Code personal memory (the per-project memory/ directory) into
tool-agnostic reference docs, so other AI CLIs that auto-load AGENTS.md (Codex
primarily; the architecture transfers to Cursor and others) working in the same
directory read the same user profile and collaboration preferences instead of being
blind to them.
Key features:
references/ + CLAUDE.md-inline + AGENTS.md-symlink, designed around
what each tool actually auto-loads (plain-text pointers are on-demand in both tools)codex verification, and memory cleanupcodex directlyExample usage:
/daymade-claude-code:claude-migrate-memory-to-doc
migrate my memory — Codex doesn't know who I am
my memory is locked to Claude Code, make it tool-agnostic
memory has grown bloated with content that should live in docs
turn off auto memory for this project and move everything into docs
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:docx-creator)
A thin incremental layer over minimax-skills:minimax-docx — not a document
engine itself. Adds the layer the underlying OpenXML engine doesn't ship: a
verified markdown-to-docx generator for Chinese formal documents, the
alignment-layering rule that stops justified text from stretching 甲方/乙方
info blocks and signature blocks into garbage, per-list numbering restart,
CJK font dual-slot setup, and a mandatory LibreOffice-to-PDF-to-PNG visual
verification chain (qlmanage thumbnails are banned — they hide exactly the
bugs that matter).
Key features:
minimax-docx; this skill owns Chinese formal-document typography rules and the CLI-vs-C# routing decisionscripts/Program.cs) for contracts, agreements, and 公文 with 甲乙方 info blocks, numbered clauses, signature blocks, and tablesExample usage:
/daymade-docs:docx-creator
生成 Word 文档
写合同 docx
把 markdown 转成 Word
give me a Word file for this labor contract
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:read-docx-review)
Read a reviewed docx (Word/WPS) and extract every comment and tracked change into a per-item adjudication table (markdown or JSON): who commented, on which paragraph, what they said, plus revision-mode insertions and deletions. Revision-aware engine on the OpenXML SDK, so content inserted via track changes (w:ins) and paragraph-level deletions (w:del) that a bare python-docx read silently misses are captured. Read-only — never mutates the reviewer's file.
Key features:
Example usage:
提取这份 docx 的批注
对方批注完的合同回来了，读一下审阅意见
读一下修订，谁批了什么、批在哪
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:claude-code-hooks)
How to write, test, register, and debug Claude Code hooks — PreToolUse / PostToolUse / SessionStart / Stop Bash guards that enforce a rule the model would otherwise talk itself past. A prose rule in CLAUDE.md is a suggestion the completion drive can override; a hook is a wall.
Key features:
bash -n + real-JSON end-to-end testing before registering, version-controlled source with installer-owned recovery, and per-profile registration convergence with human-confirmation release gatespython3 -c blockscripts/test_hook.sh)Example usage:
/daymade-claude-code:claude-code-hooks
make a hook that blocks git push --force
turn this rule I keep forgetting into a hard gate
my hook is false-blocking a healthy command
debug this hook that's poisoning my session
Install:
claude plugin install daymade-macos@daymade-skills(suite-only — invoked asdaymade-macos:macos-watchdog)
Design, deploy, and discipline macOS launchd watchdogs — LaunchAgents / LaunchDaemons that detect a recurring problem and auto-remediate it, without the watchdog itself becoming the disturbance. Distilled from 15 production watchdogs and their incident history.
Key features:
macos-permissions, and stop semantics (unload is deprecated and gets resurrected by RunAtLoad — bootout/bootstrap/disable only)watchdog-cooldown.sh (source-able escalating cooldown + manual pause state machine), new-launchagent.sh (idempotent installer with validation), and an annotated plist templateExample usage:
make a launchd watchdog that reconnects my VPN when it drops
my watchdog keeps re-launching an app I quit — make it stop
audit my existing LaunchAgents for noise and crash loops
Install:
claude plugin install daymade-macos@daymade-skills(suite-only — invoked asdaymade-macos:macos-load-doctor)
Find what is making the Mac slow or hot, attribute it to an owner, and act within the shared-machine boundary. Load average first, then a three-reading process census, parent-chain attribution, and shape classification — the skill's job is a correct, evidence-backed attribution, not process cleanup.
Key features:
vm.loadavg before any individual debuggingload_census.sh) — children-per-parent (the leak that no orphan reaper will ever clean), cumulative CPU time (the busy loop %CPU dilutes), instantaneous %CPU (the active storm)Example usage:
everything on this Mac is slow and every agent's hooks are timing out
the machine is burning hot and I can't tell why
hundreds of node processes — where are they coming from?
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:devils-advocate)
Structured devil's-advocate pressure-testing of an investment thesis against user-supplied evidence materials — evidence-anchored dissent, not role-played contrarianism (every counterpoint must cite, verbatim).
Key features:
Example usage:
/daymade-financial:devils-advocate
stress-test my bull case on <stock> against these three research reports
what would have to be true for this thesis to be wrong?
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:daymade-sector-research)
A-share (Chinese market) sector research workflow: Top-N gainers across all sector constituents, announcement-window retrieval, and market-sentiment judgment with graded evidence — executed by an Agent Team with fresh-context adversarial verification. All public, no-login data sources.
Key features:
Example usage:
/daymade-financial:daymade-sector-research
医药行业今天的 Top 10 标的
这些标的最近一个月发过哪些公告
判断医药板块现在的市场情绪
Drive the logged-in Kimi desktop app through computer-use to query its built-in company, financial, market, academic, and legal data plugins without separate API keys. Results stay source-labeled and must be independently checked before they become load-bearing data. Its separate Chat Deep Research route exports a full report and attachments.
Key features:
Example usage:
/kimi-use
用 Kimi 查这家公司的股东和最新财务数据，并逐项标明来源
操作 Kimi 客户端，用已安装插件核对这组市场数据
Install:
claude plugin install tibo-reset-codex@daymade-skills
查询重置公告，核实多个 Pro 账号的剩余额度与备用重置；有时间依据时预测下一轮，记录预测、核验与撤回，供后续判断调整。
ChatGPT 什么时候重置额度
几个 Pro 账号都用完了吗，哪个还有额度
banked reset 到了吗
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:prior-work-retrieval)
When the user explicitly asks to reuse or find earlier work, verify successful code, decisions, Skills/SOPs, meetings, WeChat archives, documents, and conversation history before producing. References to the current tests, README, files, implementation, or behavior do not trigger retrieval. Every real run leaves an auditable reuse/adapt/reject receipt; zero ranked hits never become an absence claim.
Key features:
Example usage:
/daymade-claude-code:prior-work-retrieval
we solved this before; find the code and successful path before changing anything
don't rebuild the workflow until you have checked our existing Skills and history
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:lark-cli-router)
Route Feishu/Lark/Doubao requests to the matching guide embedded in the installed
lark-cli, instead of loading every Lark domain Skill into the model catalog.
Tencent IMA stays separate under ima-skill.
Key features:
lark-cli skills listlark-cli skills readExample usage:
读取这个飞书文档
查一下这条妙记的逐字稿
lark-cli says the user identity is missing a scope
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:claude-code-ping-start-5h-quota)
Schedule a one-shot local timer (macOS) that sends one minimal claude -p message
right after your subscription quota resets, so the next 5-hour usage window starts
counting while you sleep instead of when you wake up. Dual redundancy: the detached
ping survives closing the session, and a live session is re-woken to verify the
ok receipt.
Key features:
caffeinate -is holds off idle sleep during the wait; independent ps read-back verifies the timer chainpkill cancel command (live-verified) — never kill by sleep secondsExample usage:
还有 1 小时 20 分钟重置额度，帮我定时戳一下 Claude
我去睡觉了，额度重置后帮我 ping 一下开启新窗口
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:agent-web-search-setup)
A reseller or relay that proxies Claude or Codex to another cloud (Vertex AI,
Bedrock, or an OpenAI-compatible layer) cannot execute Anthropic's server-side
web_search and web_fetch. The request still succeeds, so nothing errors — the
model simply gets no results and reports that recent things do not exist, which
reads as "the model got stupid" rather than as a broken tool. This skill probes
the endpoint to establish which built-in tools are actually dead, removes those
from the client so the model stops reaching for them, installs a client-executed
replacement it can really call, and proves it with a live query. It configures
the agent; it is not itself a search engine.
Key features:
diagnose.py (standard library only, runs before anything is installed) reads the shape of the reply rather than its text: server_tool_use means the tool really ran, a plain tool_use means it was handed back and never executed, HTTP 400 means the translation layer cannot express the tool at all--api openai asks the same question of the OpenAI Responses API, where a working endpoint returns a web_search_call item, and --api both handles a relay serving both shapesinconclusive, because every one of them would otherwise have the caller delete a working toolpermissions.allow in a project's shared .claude/settings.json is dropped in any workspace whose trust dialog has not been accepted, and the client says so on stderr alone — so an agent reading --output-format json sees a refusal with no reason, and the model tells the user it is waiting for permissiontoolu_vrtx_ Vertex AI, toolu_bdrk_ Bedrock), which is why switching models on the same endpoint does not helpWebSearch still on the menu, a model made 21 calls to it and 0 to an installed alternative whose description said the built-in was brokenExample usage:
搜索返回空的，模型说这东西不存在
我用的是中转站的 key，联网搜索好像坏了
give this agent internet access — web search does nothing
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:tech-selection)
A checklist for choosing between technologies — library, framework, storage, data format, model, build-vs-buy, architecture. Runs before committing, not after. The criteria are filters, not sorters: they kill candidates that violate a principle, and survivors are decided by business-result anchoring, never by ranking. When two or more candidates survive, the output is candidates
Key features:
unknown is not pass — a candidate carrying an unverified axis does not enter the survivor setExample usage:
用哪个框架？React 还是 Vue？
这个爬虫用 requests 还是 playwright
数据存哪里？SQLite 还是 Postgres？
要不要自建，还是用现成的？
别闭门造车，先看看有没有现成的
which library should we use for this?
build or buy — evaluate the options
Install:
claude plugin install daymade-codex@daymade-skills(suite-only — invoked asdaymade-codex:codex-1m-context-window-setup)
Configure the shared Codex CLI/Desktop base config for the largest context window the selected model actually declares, up to a one-million-token request. The Skill discovers the live model contract, explains the familiar ~258K default, applies a 60% auto-compaction threshold, preserves unrelated TOML, backs up changed bytes, and rolls back if Codex strict-config validation fails.
Key features:
doctor, transactional apply, and drift-detecting verify modesmodel_context_window and model_auto_compact_token_limitExample usage:
/daymade-codex:codex-1m-context-window-setup doctor
configure Codex CLI and Desktop for the largest verified context window
verify that my long-context setup still matches the selected model
Install:
claude plugin install daymade-codex@daymade-skills(suite-only — invoked asdaymade-codex:interaction-design-board)
Generate several genuinely different HTML interaction architectures for one bounded product surface while holding business facts and the current design language constant. The bundled builder combines the self-contained candidates into a keyboard-accessible Design Board with Focus/Compare modes, structured task feedback, remix requests, and a content-bound approval handoff. Product code stays untouched until the user has operated and approved a direction.
Key features:
Example usage:
/daymade-codex:interaction-design-board
show me three clickable ways to organize this review workflow before changing the app
the visual style is fine, but compare queue-first, object-first, and ledger-first interactions
use a Design Board so I can try each progressive-disclosure option and request a remix
📚 Documentation: See interaction-design-method.md and board-contract.md.
Glue logos, avatars, or stickers onto moving objects in a video clip so they follow the motion frame by frame — then export MP4 + GIF. Semi-supervised tracking: you read boxes off grid sheets, OpenCV CSRT carries them, and re-anchors or hand-set keyframes take over where trackers die (shot changes, walk-toward-camera scale blowups, long smooth walkaways).
When to use:
Key features:
Example usage:
"把这段视频里三只猫的头分别换成这三个 logo，做成梗图视频和 GIF"
"Cover the CEO's face with our competitor's logo in this keynote clip"
"Turn 6:30-6:50 of this bilibili video into a GIF with my avatar on the main character"
📚 Documentation: See meme-creator/SKILL.md and the bundled references/ for the tracking playbook and the asset-binding gate.
Requirements: ffmpeg; uv (bundled Python scripts carry inline dependencies). yt-dlp only when downloading from a URL.
Install:
claude plugin install data-visualization-discipline@daymade-skills
Decides whether a chart should exist and how it should be drawn before any code is written: the one conclusion each chart must support, mean vs. median, whether segments may be aggregated, stacked vs. line, dual axes, color count and cross-chart color identity, and how many columns a table keeps. Medium-agnostic — HTML reports, React/Vue dashboards, native PowerPoint charts, matplotlib/plotly/ECharts/D3.
Key features:
Five stages (intent → data validity → form → encoding → delivery): the first four end with a handoff check, the fifth is the delivery gate
A nine-item delivery gate, including a masked-render self-test and painted-span proportion checks
Reviews existing charts ("what is wrong with these charts", "why do the legends disagree")
data-visualization-discipline — full instructions. Pairs with report-with-html, which owns the report page itself.
Install:
claude plugin install report-with-html@daymade-skills
Produces the reader-facing HTML artifact — report, dashboard, architecture or journey view, data browser, or review workbench — built around the reader's question, with every number traceable and the page verified in a real browser before delivery.
Key features:
Skeleton chosen by the reader's question (how it works / what to decide / why it happened / which is better / inspect a collection)
Decision-card contract, reusable interaction components, and a warm-paper starter template
Delivery gate script with a masked (text-free) render, segment cropping for full-page review, and a zero-context independent reader review
report-with-html — full instructions. Load data-visualization-discipline for chart judgment.
Requirements: Google Chrome or Chromium; uv; Python 3.10+.
Install:
claude plugin install skill-evolver@daymade-skills
Daymade Skill Evolver assembles an improvement workflow for a specified existing Skill, using existing Skills for evidence gathering, mechanism trade-offs and authorized implementation. Reuse successful and failed work, inspect relevant competitor source, and verify changes through the target's actual consumer. Keep the current design when the evidence supports no change.
When to use: Improve a named Skill from work history, saved research or competitor mechanisms. Use Skill Creator for routine creation and editing.
Example usage:
Use skill-evolver to improve my batch-export Skill. Preserve successful exports,
investigate the interrupted-batch failure, and implement the justified repair.
Requirements: A host with access to the target source and the tools needed to edit and verify it. Resolve, load and execute available public stage Skills; local private integrations are optional. Missing required sources or data pause only dependent work.
Dependencies: This plugin ships only the orchestrator. For a missing executor, report its supported install package from the public catalog or an explicit fallback with its evidence limits.
📚 Documentation: skill-evolver.
Install:
claude plugin install app-evolver@daymade-skills
Improve an existing application from its current code and product contract, authorized successful/failed work, and competitor mechanisms. Decide what fits, implement the authorized increment and verify the actual user result.
When to use: Take the useful parts of a competitor, improve an app from past
experience, or coordinate improvements across apps with separate contracts.
Use skill-evolver when the target is a Skill.
Use app-evolver to improve my document app from this competitor. Preserve source
provenance, implement the justified improvement and verify failure recovery.
📚 Documentation: app-evolver.
Reverse Skill composition: Specialist selection and handoff.
Want to see all demos in one place with click-to-enlarge functionality? Check out our interactive demo gallery or browse the demos directory.
Use github-ops for verified PR, issue, Actions, repository, organization-access, and API operations. Use github-review-pr when a maintainer needs a current-base code review, ownership decision, or review-gated repair/landing for one contributor PR or the complete open PR queue. An author can explicitly request the same independent read-only lens for their own contribution; implementation and authorized fixes remain with github-contributor.
Combine doc-to-markdown for document conversion and mermaid-tools for diagram generation to create comprehensive documentation. Use llm-icon-finder to add brand icons.
Use deep-research to produce format-controlled research reports with evidence tables and citations. Combine with fact-checker to validate claims or with twitter-reader for social-source collection.
Use tech-selection when choosing between technologies — library, framework, storage, data format, model, build-vs-buy, or architecture. It runs a gated checklist that filters candidates by business-result anchoring, prior-art inventory, and probe-based evidence, then returns candidates + trade-offs + a recommendation when multiple survive. Combine with deep-research when the landscape itself needs research before a decision can be made.
Use competitors-analysis to discover, persist, update, and analyze competitor repositories with evidence-based source citations. Combine with deep-research when the landscape also needs broader market, pricing, or narrative research.
Use pdf-creator to convert markdown to print-ready PDFs with proper Chinese font support for formal documents and reports.
Use teams-channel-post-writer to share knowledge and statusline-generator to track costs while working.
Use the current host's native communication tools for reachable agents and sessions. Use peer-message only for uncovered local targets, cross-product/script transport, or coordination evidence that needs verification. Follow its routing contract; a denied or Held message is not a reason to switch transports.
Use repomix-unmixer to extract and validate repomix-packed skills or repositories. Use repomix-safe-mixer to package codebases securely, automatically detecting and blocking hardcoded credentials before distribution.
Use skill-creator (see Essential Skill section above) to build, validate, and package your own Claude Code skills following best practices. Use skill-evolver for a research-driven improvement round on a specified existing Skill, from relevant evidence and trade-offs to authorized changes and result verification.
Use ppt-creator to generate professional slide decks with data visualizations, structured storytelling, and complete PPTX output for pitches, reviews, and keynotes. Use slides-creator for narrative-first slide design — it guides you through the ABCDEFG storytelling framework, collects your original content first, then delegates visual generation to baoyu-slide-deck. Perfect when you have existing articles, transcripts, or talks that need to become visual slides.
Use video-comparer to analyze compression results, evaluate codec performance, and generate interactive comparison reports. Combine with youtube-downloader to compare different quality downloads.
Use youtube-downloader t
Truncated — view the full README on GitHub.
47 followers · starred May 2026
138 followers · starred May 2026
106 followers · starred Feb 2026
22 followers · starred Nov 2025
Professional Claude Code skills marketplace featuring production-ready skills for enhanced development workflows.
Python
1,458
1,397 commits
updated Oct 14, 2026
Professional Claude Code skills marketplace featuring production-ready skills for enhanced development workflows.
⭐ Start here if you want to create your own skills!
The skill-creator is the meta-skill that enables you to build, validate, and package your own Claude Code skills. It's the most important tool in this marketplace because it empowers you to extend Claude Code with your own specialized workflows.
This is a production-hardened fork of Anthropic's official skill-creator, born from building real skills and hitting every wall the official version doesn't warn you about.
The official skill-creator tells you what to build. Ours also tells you what not to try — and why.
| You're trying to... | Official | This Fork |
|---|---|---|
| Research before building | "Check available MCPs" | Ordered search protocol with decision matrix: Adopt / Extend / Build |
| Create a skill interactively | Prose-based instructions | Structured AskUserQuestion checkpoints — user never loses context |
| Avoid common mistakes | No guidance | Cache edit warnings, prerequisite checks, security scan gate |
| Know the architecture options | Not mentioned | Inline vs Fork decision guide with examples (choosing wrong silently breaks your skill) |
| Validate before shipping | Basic YAML check | Expanded structural validator plus provenance-checked old-vs-new capability audit; packaging re-verifies the completed review instead of trusting a marker |
| Match validation cost to risk | One full create-test-review loop | Risk tier and evaluation spend are separate: targeted checks first; paired benchmarks require explicit authorization or a decision-bearing evidence plan plus opt-in |
| Catch security issues | No tooling | security_scan.py with gitleaks integration — hard gate before packaging |
| Learn from real failures | No failure cases | Battle-tested methodology with documented failure patterns and gotchas |
| Distill past conversations safely | Not covered | Explicit local manifest, message-level time window, redaction, opaque source IDs, ignored .enrich/ staging, and manual promotion into references/scripts |
| Turn approved artifacts into skill behavior | Not covered | Artifact-corpus distillation workflow: script-measured pattern extraction (≥3-artifact threshold), cataloging≠distillation gate, invariant-vs-register boundary, independent completeness audit |
| Survive concurrent sessions on one skill repo | Not covered | git-ref baselines, re-read-before-write, HEAD check before commit, own-paths-only staging, one-bump-per-outcome versioning |
| Ground knowledge skills in evidence | General advice | Authority ladder from real calls and machine-readable specs through production code, plus executable-example smoke checks and evidence-boundary rules |
| Have both installed at once | Coin flip — the two descriptions are near-identical | Detects the clash on trigger and offers a one-command, reversible SessionStart routing hook (only ever installed when both coexist); the official plugin stays usable by explicit request |
| Your own skill collides with an installed plugin | Not covered | generate_supersede_kit.py stamps the same conditional routing kit into your skill, plus a measured precedence decision guide (rename → description tiebreaker → hook → disable) |
Full methodology: skill-creator/references/skill-development-methodology.md
In Claude Code (in-app):
/plugin marketplace add daymade/claude-code-skills
Then:
From your terminal (CLI):
claude plugin marketplace add https://github.com/daymade/claude-code-skills
# Marketplace name: daymade-skills (from marketplace.json)
claude plugin install daymade-skill@daymade-skills
After installing skill-creator, simply ask Claude Code:
"Create a new skill called my-awesome-skill in ~/my-skills"
"Validate my skill at ~/my-skills/my-awesome-skill"
"Package my skill at ~/my-skills/my-awesome-skill for distribution"
Claude Code, with skill-creator loaded, will guide you through the entire skill creation process - from understanding your requirements to packaging the final skill.
📚 Full documentation: daymade-skill/skill-creator/SKILL.md
📝 Initialize New Skill
✅ Validate Skill Structure
📦 Package Skill for Distribution
/plugin marketplace add daymade/claude-code-skills
Then:
macOS/Linux:
curl -fsSL https://raw.githubusercontent.com/daymade/claude-code-skills/main/scripts/install.sh | bash
Windows (PowerShell):
iwr -useb https://raw.githubusercontent.com/daymade/claude-code-skills/main/scripts/install.ps1 | iex
Add the marketplace:
claude plugin marketplace add https://github.com/daymade/claude-code-skills
Marketplace name is daymade-skills (from marketplace.json). Use @daymade-skills when installing plugins.
Do not use the repo path as a marketplace name (e.g. @daymade/claude-code-skills will fail).
In Claude Code, use /plugin ... slash commands. In your terminal, use claude plugin ....
Essential Skill (recommended first install):
# skill-creator ships inside the daymade-skill suite
claude plugin install daymade-skill@daymade-skills
Documentation Suite (shared namespace for document workflows):
claude plugin install daymade-docs@daymade-skills
This suite exposes related skills under one namespace, including:
/daymade-docs:docs-router
/daymade-docs:doc-to-markdown
/daymade-docs:mermaid-tools
/daymade-docs:pdf-creator
/daymade-docs:ppt-creator
/daymade-docs:docs-cleaner
/daymade-docs:excel-automation
/daymade-docs:read-docx-review
These skills ship as a bundle — there are no separate single-skill plugins. All documentation skills live under daymade-docs/ and install together from the suite.
The router handles automatic selection. The specialist commands remain available for manual use;
ppt-creator is manual-only. New presentation creation uses deck-creator when installed.
Apple Platform Suite (shared namespace for macOS and iOS development/operations):
claude plugin install daymade-macos@daymade-skills
/daymade-macos:capture-screen
/daymade-macos:developing-ios-apps
/daymade-macos:macos-cleaner
/daymade-macos:macos-load-doctor
/daymade-macos:macos-permissions
/daymade-macos:macos-watchdog
These skills are bundle-only under daymade-macos.
Codex Suite (shared namespace for Codex workstation setup, assisted coding, and visual exploration):
claude plugin install daymade-codex@daymade-skills
Codex CLI and Desktop users can install the same suite through Codex's plugin marketplace:
codex plugin marketplace add daymade/claude-code-skills
codex plugin add daymade-codex@daymade-skills
/daymade-codex:codex-image-gallery
/daymade-codex:local-codex
/daymade-codex:design-style-picker
/daymade-codex:interaction-design-board
/daymade-codex:codex-1m-context-window-setup
These skills are bundle-only under daymade-codex.
Claude Code Operations Suite (shared namespace for Claude Code power-user workflows):
claude plugin install daymade-claude-code@daymade-skills
This suite bundles the skills that extend Claude Code itself — cross-project prior-work retrieval across code, docs, Skills, meetings, WeChat archives, and conversation history; fast local conversation discovery across Claude Code and Codex; session recovery; CLAUDE.md tuning; version-synced Lark CLI routing; troubleshooting; statusline configuration; export repair; marketplace development and suite consolidation; terminal screenshot rendering; usage analysis; and multi-provider model switching:
/daymade-claude-code:claude-code-ops-router
/daymade-claude-code:local-conversation-history
/daymade-claude-code:read-claude-code-history
/daymade-claude-code:read-codex-history
/daymade-claude-code:continue-claude-code-work
/daymade-claude-code:continue-codex-work
/daymade-claude-code:claude-skills-troubleshooting
/daymade-claude-code:claude-md-progressive-disclosurer
/daymade-claude-code:statusline-generator
/daymade-claude-code:fixing-claude-export-conversations
/daymade-claude-code:marketplace-dev
/daymade-claude-code:terminal-screenshot
/daymade-claude-code:claude-usage-analyst
/daymade-claude-code:claude-switch-models-setup
/daymade-claude-code:read-claude-web-conversation
/daymade-claude-code:claude-migrate-memory-to-doc
/daymade-claude-code:claude-code-hooks
/daymade-claude-code:prior-work-retrieval
/daymade-claude-code:lark-cli-router
/daymade-claude-code:claude-code-ping-start-5h-quota
/daymade-claude-code:agent-web-search-setup
/daymade-claude-code:tech-selection
Installed names render as daymade-claude-code:<skill> under a single shared namespace. These skills are bundle-only — install the suite to get all members.
Financial Data Suite (shared namespace for investment-research and financial data workflows):
claude plugin install daymade-financial@daymade-skills
This suite bundles the skills that fetch and analyze financial data — Bigdata.com (RavenPack) structured financials and sentiment, US equity fundamentals via yfinance, Gangtise (岗底斯) OpenAPI research suite orchestration, A-share news and policy aggregation, A-share pharmaceutical sector daily reporting, structured devil's-advocate pressure-testing of investment theses, and adversarial due diligence on inflated benchmark claims:
/daymade-financial:financial-router
/daymade-financial:bigdata-skill
/daymade-financial:financial-data-collector
/daymade-financial:gangtise-copilot
/daymade-financial:ashare-news-fetcher
/daymade-financial:daymade-sector-research
/daymade-financial:pharma-daily-report
/daymade-financial:devils-advocate
/daymade-financial:benchmark-due-diligence
Installed names render as daymade-financial:<skill> under a single shared namespace. These skills are bundle-only — install the suite to get all members.
Install Other Skills:
# Research-driven improvement of a specified existing Skill
claude plugin install skill-evolver@daymade-skills
# GitHub operations
claude plugin install github-ops@daymade-skills
# Teams communication
claude plugin install teams-channel-post-writer@daymade-skills
# Local Claude/Codex agent messaging
claude plugin install peer-message@daymade-skills
# Ghostty terminal session snapshot/restore
claude plugin install ghostty-use@daymade-skills
# Repomix extraction
claude plugin install repomix-unmixer@daymade-skills
# AI/LLM icons
claude plugin install llm-icon-finder@daymade-skills
# CLI demo generation
claude plugin install cli-demo-generator@daymade-skills
# Cloudflare diagnostics
claude plugin install cloudflare-troubleshooting@daymade-skills
# UI design system extraction
claude plugin install ui-designer@daymade-skills
# YouTube video/audio downloading
claude plugin install youtube-downloader@daymade-skills
# Secure repomix packaging
claude plugin install repomix-safe-mixer@daymade-skills
# Full audio suite (ASR + transcript correction + meeting minutes + TTS)
claude plugin install daymade-audio@daymade-skills
# Video comparison and quality analysis
claude plugin install video-comparer@daymade-skills
# QA testing infrastructure with autonomous execution
claude plugin install qa-expert@daymade-skills
# Prompt optimization using EARS methodology
claude plugin install prompt-optimizer@daymade-skills
# CCPM skill registry search and management
claude plugin install daymade-skill@daymade-skills
# Promptfoo LLM evaluation framework
claude plugin install promptfoo-evaluation@daymade-skills
# Twitter/X content fetching
claude plugin install twitter-reader@daymade-skills
# Skill quality review and improvement
claude plugin install daymade-skill@daymade-skills
# GitHub contribution strategy
claude plugin install github-contributor@daymade-skills
# Windows Remote Desktop / AVD connection diagnosis
claude plugin install windows-remote-desktop-connection-doctor@daymade-skills
# Product analysis and optimization
claude plugin install product-analysis@daymade-skills
# Scrapling CLI extraction and troubleshooting
claude plugin install scrapling-skill@daymade-skills
# Tencent IMA knowledge base companion and installer
claude plugin install ima-copilot@daymade-skills
# Export Douban (豆瓣) book/movie/music/game collections to CSV
claude plugin install douban-skill@daymade-skills
# Terraform operational traps and multi-environment reliability patterns
claude plugin install terraform-skill@daymade-skills
# Evaluate any LLM endpoint across speed, concurrency, protocol, and quality
claude plugin install llm-eval-harness@daymade-skills
# Video/GIF memes with motion-tracked image overlays
claude plugin install meme-creator@daymade-skills
Standalone plugins can be installed independently; suite members install together with their suite.
For Chinese users: We highly recommend using CC-Switch to manage Claude Code API provider configurations.
CC-Switch enables you to:
Setup: Download from Releases, install, add your API configs, and switch via UI or system tray.
For full documentation in Chinese, see README.zh-CN.md.
GitHub operations through gh CLI and GitHub APIs, with explicit targets, impact previews, and independent state verification.
When to use:
Key features:
🎬 Live Demo
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:docs-router)
Selects the bundled specialist for document conversion, Word and PDF production,
Mermaid images, macOS Excel automation, scanned PDFs, DOCX review extraction, or
documentation cleanup. It reads the selected specialist's full instructions and
required references. Specialist slash commands remain available for manual use.
New presentation creation uses deck-creator when installed; ppt-creator is manual-only.
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:doc-to-markdown)
Converts documents to markdown with Windows/WSL path handling and PDF image extraction.
When to use:
Key features:
🎬 Live Demo
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:mermaid-tools)
Extracts Mermaid diagrams from markdown and generates high-quality PNG images.
When to use:
Key features:
🎬 Live Demo
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:claude-code-ops-router)
Routes plugin and Skill repair, marketplace work, statusline, model profiles and source sync, 1M context-window repair, usage and quota timers, memory migration, and exported .txt repair to one bundled specialist. The selected Skill's full instructions are read at use time; its original slash command remains available for manual use.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:statusline-generator)
Configures Claude Code statuslines with multi-line layouts and cost tracking.
When to use:
Key features:
🎬 Live Demo
Creates educational Teams channel posts for internal knowledge sharing.
When to use:
Key features:
🎬 Live Demo
Follow the local coordination workflow for task/resource declarations, native preflight installation, recovery and receiver checks.
An experimental paired-network route adds invitations and cited answers from owner-selected documents through existing Claude/Codex hosts. It requires an operator-provided relay; the public hosted network and user-adoption claims are not part of this preview.
Install:
claude plugin install peer-message@daymade-skills
Follow the routing contract for native communication and transport gaps.
When to use:
crossSessionInbound)📚 Execution entry and detailed SOPs.
Extracts files from repomix-packed repositories and restores directory structures.
When to use:
Key features:
🎬 Live Demo
Access 100+ AI model and LLM provider brand icons from lobe-icons library.
When to use:
Key features:
🎬 Live Demo
Generate professional animated CLI demos and terminal recordings with VHS automation.
When to use:
Key features:
🎬 Live Demo
Investigate and resolve Cloudflare configuration issues using API-driven evidence gathering.
When to use:
Key features:
🎬 Live Demo
Extract design systems from reference UI images and generate implementation-ready design prompts.
When to use:
Key features:
🎬 Live Demo
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:ppt-creator)
Create persuasive, audience-ready slide decks from topics or documents with data-driven charts and dual-format PPTX output.
When to use:
Key features:
🎬 Live Demo
Download YouTube videos and audio using yt-dlp with robust error handling and automatic workarounds for common issues.
For spoken-content text, reuse existing transcripts first, then fetch original subtitles; download audio for ASR only when those checked routes miss. File-download requests keep the existing workflow.
When to use:
Key features:
🎬 Live Demo
Safely package codebases with repomix by automatically detecting and removing hardcoded credentials before packing.
When to use:
Key features:
🎬 Live Demo
Coming soon
Install:
claude plugin install daymade-audio@daymade-skills(suite-only — invoked asdaymade-audio:transcript-fixer)
Correct speech-to-text (ASR/STT) errors with a Stage 1 dictionary pre-filter, a required Native AI whole-transcript pass, and an audio-backed human gate for unresolved wording, names, numbers, and entities.
When to use:
Key features:
Example workflow:
# Initialize and add corrections
uv run scripts/fix_transcription.py --init
uv run scripts/fix_transcription.py --add "错误词" "正确词" --domain general
# Run full correction pipeline
uv run scripts/fix_transcription.py --input meeting.md --stage 3
# Open only this transcript's unresolved items and listen before deciding
uv run scripts/review-dashboard/server.py --file "/absolute/meeting.md"
# Read the human verdicts back; pending_total must be 0 before final delivery
uv run scripts/fix_transcription.py \
--list-review --review-file "/absolute/meeting.md" \
--review-status all --json
📚 Documentation: See daymade-audio/transcript-fixer/references/ for workflow guides, SQL queries, troubleshooting, best practices, team collaboration, and API setup.
Requirements: Python 3.10+ and uv. Native AI correction uses the active agent; an external API key is needed only for the optional agent-less API route.
Compare two videos and generate interactive HTML reports with quality metrics and frame-by-frame visual comparisons.
When to use:
Key features:
Example usage:
# Basic comparison
python3 scripts/compare.py original.mp4 compressed.mp4
# Custom output and frame interval
python3 scripts/compare.py original.mp4 compressed.mp4 -o report.html --interval 10
# Batch processing
for original in originals/*.mp4; do
compressed="compressed/$(basename "$original")"
output="reports/$(basename "$original" .mp4).html"
python3 scripts/compare.py "$original" "$compressed" -o "$output"
done
🎬 Live Demo
Coming soon
📚 Documentation: See video-comparer/references/ for quality metrics interpretation, FFmpeg commands, and configuration options.
Requirements: Python 3.8+, FFmpeg/FFprobe (install via brew install ffmpeg, apt install ffmpeg, or winget install ffmpeg)
Establish world-class QA testing processes with autonomous LLM execution, Google Testing Standards, and OWASP security best practices.
When to use:
Key features:
Example usage:
# Initialize QA project (creates full infrastructure)
python3 scripts/init_qa_project.py my-app ./
# Calculate quality metrics and gates status
python3 scripts/calculate_metrics.py tests/TEST-EXECUTION-TRACKING.csv
# For autonomous execution, copy master prompt from:
# references/master_qa_prompt.md → paste to LLM → execute the project's approved test plan
🎬 Live Demo
Coming soon
📚 Documentation: See qa-expert/references/ for:
master_qa_prompt.md - Autonomous execution entrygoogle_testing_standards.md - AAA pattern, coverage thresholds, OWASP testingday1_onboarding.md - Onboarding timeline for new QA engineersground_truth_principle.md - Preventing doc/CSV sync issuesllm_prompts_library.md - 30+ ready-to-use QA promptsRequirements: Python 3.8+
💡 Innovation: The autonomous execution capability (via master prompt) enables LLM to execute entire test suites 100x faster than manual execution, with zero human error in tracking. Perfect for third-party QA handoffs - just provide the master prompt and they can start testing immediately.
Transform vague prompts into precise, well-structured specifications using EARS (Easy Approach to Requirements Syntax) - a methodology created by Rolls-Royce for converting natural language into testable requirements.
Methodology inspired by: 阿星AI工作室 (A-Xing AI Studio), which pioneered combining EARS with domain theory grounding for practical prompt enhancement.
When to use:
Key features:
Example usage:
# Before (vague)
"Build me a password reset feature"
# After EARS transformation (7 atomic requirements)
1. When user clicks "Forgot Password", the system shall display email input field
2. When user submits valid email, the system shall send password reset link valid for 1 hour
3. When user clicks reset link, the system shall verify token has not expired
4. When token is valid, the system shall display password creation form requiring minimum 12 characters, 1 uppercase, 1 number, 1 special character
5. When user submits new password meeting requirements, the system shall hash password with bcrypt and invalidate reset token
6. When user attempts password reset more than 3 times in 1 hour, the system shall block further attempts for 1 hour
7. If reset token has expired, the system shall display error message and option to request new link
# Enhanced with domain theories
- Zero Trust Architecture (verify at each step)
- Defense in Depth (rate limiting + token expiration + password complexity)
- Progressive Disclosure (multi-step UX flow)
# Full prompt includes Role, Skills, Workflows, Examples, Formats
🎬 Live Demo
Coming soon
📚 Documentation: See prompt-optimizer/references/ for:
ears_syntax.md - Complete EARS patterns and transformation rulesdomain_theories.md - 40+ theories mapped to domains with selection guidanceexamples.md - Full transformation examples with before/after comparisons💡 Innovation: EARS methodology eliminates ambiguity by forcing explicit conditions, triggers, and measurable criteria. Combined with domain theory grounding (GTD, BJ Fogg, Gestalt, etc.), it transforms "build a todo app" into a complete specification with behavioral psychology principles, UX best practices, and concrete test cases - enabling test-driven development from day one.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:local-conversation-history)
See the Skill's usage and routing contract for history recall, interrupted work and conversation-backup health checks.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:read-claude-code-history)
Read, search, and export Claude Code history across all active config homes and
the long-term archives registered in ~/.claude/history-sources.json, without
resuming or modifying the old task.
When to use:
Key features:
read-codex-history--all-projects searches every Claude project when the project is unknown; --exclude-session skips self-matchesExample usage:
# List recent sessions for a project
python3 scripts/analyze_sessions.py list /path/to/project
# Read one Session chronologically, preserving queued human input
python3 scripts/read_claude_session.py --session <session-id> --project /path/to/project --full
# Export recent human wording grouped by Session
python3 scripts/extract_user_messages.py ./user-words --days 7 --group-by session
# Search sessions for keywords
python3 scripts/analyze_sessions.py search /path/to/project \
"ComponentName" "featureX" --from-date 2026-03-01 --to-date 2026-04-30
# Recover deleted files from the exact path printed by search
python3 scripts/recover_content.py <printed-session-path> -k DeletedComponent -o ./recovered/
# Search every project for multiple vanished job artifacts
python3 scripts/analyze_sessions.py search --all-projects \
artifact-a.html artifact-b.html \
--exclude-session <current-session-id>
# Get session statistics
python3 scripts/analyze_sessions.py stats /path/to/session.jsonl --show-files
🎬 Live Demo
Coming soon
📚 Documentation: See daymade-claude-code/read-claude-code-history/references/ for:
session_file_format.md - JSONL structure and extraction patternsworkflow_examples.md - Detailed recovery and analysis workflowsInstall:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:docs-cleaner)
Consolidate redundant documentation while preserving all valuable content.
When to use:
Key features:
🎬 Live Demo
Coming soon
Find Skills in configurable local repositories first, then approved sources and the CCPM registry. Preserve existing CCPM installation and management commands.
When to use:
Key features:
ccpm search <query>ccpm install <skill-name>@version syntaxccpm info <skill-name>Example usage:
# Search configured local repositories first (<skill-dir> is this loaded Skill)
uv run --script <skill-dir>/scripts/local_sources.py search pdf --tier owned
# Expand to the external registry when needed
ccpm search pdf # Find PDF-related skills
ccpm search "code review" # Find code review skills
# Install skills
ccpm install skill-creator # From registry
ccpm install daymade/skill-creator # From GitHub
ccpm install skill-creator@1.0.0 # Specific version
# List and manage
ccpm list # List installed skills
ccpm info skill-creator # Get skill details
ccpm uninstall pdf-processor # Remove a skill
# Install bundles
ccpm install-bundle web-dev # Install web development skills bundle
🎬 Live Demo
Coming soon
📚 Documentation: See daymade-skill/skills-search/SKILL.md for configuration, coverage and commands
Requirements: Python 3.10+, Git and uv for local search; CCPM CLI only for external registry/install/manage operations
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:pdf-creator)
Create professional PDF documents from markdown with proper Chinese typography using WeasyPrint.
When to use:
Key features:
Example usage:
uv run --with weasyprint scripts/md_to_pdf.py input.md output.pdf
🎬 Live Demo
Coming soon
📚 Documentation: See daymade-docs/pdf-creator/SKILL.md for setup and workflow details.
Requirements: Python 3.8+, pandoc (system install), weasyprint (or Chrome as fallback backend)
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:claude-md-progressive-disclosurer)
Optimize user CLAUDE.md files using progressive disclosure to reduce context bloat while preserving critical rules.
When to use:
Key features:
Example usage:
"Optimize my ~/.claude/CLAUDE.md using progressive disclosure and propose a plan."
🎬 Live Demo
Coming soon
📚 Documentation: See claude-md-progressive-disclosurer/SKILL.md.
Configure and run LLM evaluations with Promptfoo for prompt testing and model comparisons.
When to use:
Key features:
Example usage:
npx promptfoo@latest init
npx promptfoo@latest eval
npx promptfoo@latest view
🎬 Live Demo
Coming soon
📚 Documentation: See promptfoo-evaluation/references/promptfoo_api.md.
Requirements: Node.js (Promptfoo via npx promptfoo@latest)
Install:
claude plugin install daymade-macos@daymade-skills(suite-only — invoked asdaymade-macos:developing-ios-apps)
Build, configure, and debug iOS apps with XcodeGen, SwiftUI, and Swift Package Manager.
When to use:
project.ymlKey features:
Example usage:
xcodegen generate
xcodebuild -destination 'platform=iOS Simulator,name=iPhone 17' build
🎬 Live Demo
Coming soon
📚 Documentation: See developing-ios-apps/references/xcodegen-full.md.
Requirements: macOS + Xcode, XcodeGen
Fetch Twitter/X post and article content. Single-post text goes through the
fxtwitter mirror API (login-free, key-free, full long-form body); X Articles
with images go through fetch_article.py; the Jina API path remains as a
fallback with a known intermittent-ban risk.
When to use:
Key features:
tweet.text body including note_tweet long-formExample usage:
# Single post text (preferred): fxtwitter mirror, no key needed
curl -sS "https://api.fxtwitter.com/USER/status/TWEET_ID" \
| python3 -c "import json,sys; t=json.load(sys.stdin)['tweet']; print(t['text'])"
# X Article with images (full mode)
uv run --with pyyaml python scripts/fetch_article.py \
"https://x.com/USER/status/TWEET_ID" ./Clippings
# Jina fallback (intermittent — see SKILL.md risk note)
export JINA_API_KEY="your_api_key_here"
curl "https://r.jina.ai/https://x.com/USER/status/TWEET_ID" \
-H "Authorization: Bearer ${JINA_API_KEY}"
# Batch fetch multiple tweets
scripts/fetch_tweets.sh \
"https://x.com/user/status/123" \
"https://x.com/user/status/456"
# Fetch to file using Python script
python scripts/fetch_tweet.py https://x.com/user/status/123 output.md
🎬 Live Demo
Coming soon
📚 Documentation: See twitter-reader/SKILL.md for full details and URL format support.
Requirements:
Install:
claude plugin install daymade-macos@daymade-skills(suite-only — invoked asdaymade-macos:macos-cleaner)
The safest way to reclaim disk space on macOS. Start with targeted read-only diagnosis when a subsystem is already suspect—including Apple Content Caching—then analyze caches, application remnants, large files, and development environments only as needed.
Why macos-cleaner stands out:
Our design principles:
When to use:
Caching needs more space, an unlimited cache, or unexpectedly high ActualCacheUsedKey features:
CacheUsed from physical ActualCacheUsed, verifies peers and protected services, and uses Apple's supported limit/deactivate/flush controlsfdupes scans inside exact approved pathsWhat makes us different:
.git directories, node_modules, and build artifacts.Example usage:
# Install the Apple platform suite
claude plugin install daymade-macos@daymade-skills
# Ask Claude Code to analyze your Mac
"My Mac is running out of space, help me analyze what's using storage"
# Claude will:
# 1. Route a named suspect to targeted read-only diagnosis; scan broadly only if the source is unknown
# 2. Present categorized findings with safety levels
# 3. Explain each category (caches, remnants, large files, dev tools)
# 4. Recommend cleanup approach
# 5. Execute ONLY what you confirm
# Example analysis output:
📊 Disk Space Analysis
━━━━━━━━━━━━━━━━━━━━━━━━
Total: 500 GB
Used: 450 GB (90%)
Available: 50 GB (10%)
🟢 Safe to Clean (95 GB):
- System caches: 45 GB (apps regenerate automatically)
- Homebrew cache: 5 GB (reinstalls when needed)
- npm cache: 3 GB (safe to clear)
- Old logs: 8 GB (diagnostic data only)
- Trash: 34 GB (already marked for deletion)
🟡 Review Recommended (62 GB):
- Large downloads: 38 GB (may contain important files)
- App remnants: 8 GB (verify apps are truly uninstalled)
- Docker images: 12 GB (may be in use)
- Old .git repos: 4 GB (verify project is archived)
🔴 Keep Unless Certain (0 GB):
- No high-risk items detected
Recommendation: Start with 🟢 Safe items (95 GB), then review 🟡 items together.
🎬 Live Demo
Coming soon
📚 Documentation: See macos-cleaner/references/ for:
apple_content_caching.md - Targeted Apple Content Caching diagnosis and supported repaircleanup_targets.md - Detailed explanations of every cleanup targetmole_integration.md - How to combine scripts with Mole visual toolsafety_rules.md - Comprehensive safety guidelines and what to never deleteRequirements:
Verify factual claims in documents using web search and official sources, then propose corrections with user confirmation.
When to use:
Key features:
Example usage:
# Install the skill
claude plugin install fact-checker@daymade-skills
# Fact-check a document
"Please fact-check this section about AI model capabilities"
# Verify technical specs
"Check if these Claude model specifications are still accurate"
# Update outdated info
"Verify and update the version numbers in this documentation"
🎬 Live Demo
Coming soon
📚 Documentation: See fact-checker/SKILL.md for full workflow and claim types.
Requirements:
Review and improve Claude Code skills against official best practices with three powerful modes.
When to use:
Key features:
Example usage:
# Install the skill
claude plugin install daymade-skill@daymade-skills
# Self-review your skill
"Validate my skill at ~/my-skills/my-awesome-skill"
# Review external skill repository
"Review the skills at https://github.com/user/skill-repo"
# Auto-PR improvements
"Fork, improve, and submit PR for https://github.com/user/skill-repo"
🎬 Live Demo
Coming soon
📚 Documentation: See daymade-skill/skill-reviewer/references/ for:
evaluation_checklist.md - Complete skill evaluation criteriapr_template.md - Professional PR description templateStrategic guide for becoming an effective GitHub contributor and building your open-source reputation.
When to use:
Key features:
Contribution types explained:
Level 1: Documentation fixes (lowest barrier, high impact)
↓ (build familiarity)
Level 2: Code quality (medium effort, demonstrates skill)
↓ (understand codebase)
Level 3: Bug fixes (high impact, builds trust)
↓ (trusted contributor)
Level 4: Feature additions (highest visibility)
↓ (potential maintainer)
Example usage:
# Install the skill
claude plugin install github-contributor@daymade-skills
# Find good first issues
"Help me find projects with good first issues in Python"
# Write a high-quality PR
"Guide me through creating a PR for this bug fix"
# Build contribution strategy
"Help me plan a contribution strategy for building my GitHub profile"
🎬 Live Demo
Coming soon
📚 Documentation: See github-contributor/references/ for:
pr_checklist.md - Complete PR quality checklistproject_evaluation.md - How to evaluate projects for contributioncommunication_templates.md - Issue and PR communication templatesComplete internationalization/localization setup and auditing for UI codebases. Configure i18n frameworks, replace hard-coded strings with translation keys, ensure locale parity between en-US and zh-CN, and validate pluralization and formatting.
When to use:
Key features:
Example usage:
# Install the skill
claude plugin install i18n-expert@daymade-skills
# Setup i18n for a new project
"Set up i18n for my React app with English and Chinese support"
# Audit existing i18n implementation
"Audit the i18n setup and find missing translation keys"
# Replace hard-coded strings
"Replace all hard-coded strings in this component with i18n keys"
🎬 Live Demo
Coming soon
📚 Documentation: See i18n-expert/SKILL.md for complete workflow and architecture guidance.
Requirements:
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:claude-skills-troubleshooting)
Diagnose and resolve Claude Code plugin and skill configuration issues. Debug plugin installation, enablement, and activation problems with systematic workflows.
When to use:
Key features:
Example usage:
# Run diagnostic
python3 scripts/diagnose_plugins.py
# Batch enable missing plugins
python3 scripts/enable_all_plugins.py daymade-skills
🎬 Live Demo
Coming soon
📚 Documentation: See claude-skills-troubleshooting/SKILL.md for complete troubleshooting workflow and architecture guidance.
Requirements: None (uses Claude Code built-in Python)
Install:
claude plugin install daymade-audio@daymade-skills(suite-only — invoked asdaymade-audio:audio-router)
Selects the installed StepFun ASR, StepFun TTS, or transcript-to-minutes specialist and reads its full instructions at use time. Their original slash commands remain available manually. General audio transcription and transcript correction retain direct automatic entries.
Install:
claude plugin install daymade-audio@daymade-skills(suite-only — invoked asdaymade-audio:meeting-minutes-taker)
Transform meeting transcripts into high-fidelity, structured meeting minutes with iterative human review.
When to use:
Key features:
Example usage:
# Install the full audio suite (includes meeting-minutes-taker)
claude plugin install daymade-audio@daymade-skills
# Then provide a meeting transcript and request minutes
🎬 Live Demo
Coming soon
📚 Documentation: See daymade-audio/meeting-minutes-taker/SKILL.md for complete workflow and template guidance.
Requirements: None
Generate research reports from a durable study record: original sources, claim-level citations, rejected leads, and prior studies remain available for the next question. When several AI products or modes investigate one decision, coordinate their existing Skills and agents in parallel, preserve original outputs, and synthesize against the underlying sources and business question.
When to use:
Key features:
Example usage:
# Install the skill
claude plugin install deep-research@daymade-skills
# Then provide a report spec or template and request a deep research report
🎬 Live Demo
Coming soon
📚 Documentation: See deep-research/SKILL.md, research-asset-contract.md, and research_report_template.md.
Requirements: None
Discover, clone, update, and analyze competitor repositories with evidence-based competitive intelligence. Repository-backed findings must come from local cloned code; market-landscape claims must cite their source and volatility.
When to use:
Key features:
$HOME/workspace/competitors/{product}/file:line_number)Example usage:
# Install the skill
claude plugin install competitors-analysis@daymade-skills
# Then ask Claude to analyze a competitor
"分析竞品 https://github.com/org/repo"
"添加竞品到 flowzero 产品的竞品列表"
"看看 claude-flow-viewer 这个方向最近有哪些竞品更新"
🎬 Live Demo
Coming soon
📚 Documentation: See competitors-analysis/SKILL.md and competitors-analysis/references/ for templates.
Requirements: Git (for cloning repositories)
Diagnose and fix conflicts when using Tailscale alongside proxy/VPN tools (Shadowrocket, Clash, Surge) on macOS. Covers four independent conflict layers with specific guidance for SSH access to WSL instances.
When to use:
git push/pull fails with "failed to begin relaying via HTTP"operation not permittedKey features:
check vs accept)Example usage:
# Install the skill
claude plugin install tunnel-doctor@daymade-skills
# Then ask Claude to diagnose
"Tailscale ping works but SSH times out"
"Fix Tailscale and Shadowrocket route conflict on macOS"
"git push fails with failed to begin relaying via HTTP"
"Set up Tailscale SSH to my WSL instance"
🎬 Live Demo
Coming soon
📚 Documentation: See tunnel-doctor/references/proxy_conflict_reference.md for per-tool configuration and conflict architecture.
Diagnose Windows App (Microsoft Remote Desktop / Azure Virtual Desktop / W365) connection quality issues on macOS, with focus on transport protocol optimization (UDP Shortpath vs WebSocket fallback).
When to use:
Key features:
Example usage:
# Install the skill
claude plugin install windows-remote-desktop-connection-doctor@daymade-skills
# Then ask Claude to diagnose
"My VDI connection shows WebSocket instead of UDP, RTT is 165ms"
"Diagnose why RDP Shortpath is not working"
"Windows App transport protocol stuck on WebSocket"
🎬 Live Demo
Coming soon
📚 Documentation: See windows-remote-desktop-connection-doctor/references/ for log analysis patterns and AVD transport protocol details.
Run a scalable, evidence-driven product audit using parallel Claude Code agents and optional Codex CLI parallelization. Covers UX, API, architecture, and competitive benchmark workflows with quantified findings and priority recommendations.
When to use:
Key features:
codex availability)full, ux, api, arch, comparecompetitors-analysisExample usage:
# Install the skill
claude plugin install product-analysis@daymade-skills
# Then ask Claude for analysis
"Run product-analysis in full mode for launch audit"
"Do a UX audit and report quantified navigation findings"
"Run API audit and identify unused endpoints"
"Compare this product with our top competitors"
🎬 Live Demo
Coming soon
📚 Documentation: See product-analysis/SKILL.md and product-analysis/references/analysis_dimensions.md for dimension definitions and workflow guidance.
Requirements: Optional codex CLI (for multi-model parallel mode). Skill runs with Claude only if codex is not installed.
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:financial-data-collector)
Collect real financial data for any US publicly traded company from free public sources (yfinance). Output structured JSON with market data, historical financials (income statement, cash flow, balance sheet), WACC inputs, and analyst estimates - ready for downstream DCF modeling, comps analysis, or earnings review.
When to use:
Key features:
null with _source attribution, never default valuesExample usage:
# Install the suite
claude plugin install daymade-financial@daymade-skills
# Then ask Claude to collect data
"Collect financial data for META"
"Get financials for AAPL --years 3"
"Pull DCF inputs for NVDA"
🎬 Live Demo
Coming soon
📚 Documentation: See financial-data-collector/SKILL.md, output-schema.md, and yfinance-pitfalls.md.
Requirements: Python 3.11+, yfinance, pandas (auto-installed via uv inline dependencies).
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:excel-automation)
Create professionally formatted Excel files, parse complex .xlsm models with stdlib XML/ZIP workflows, and control Microsoft Excel windows on macOS via AppleScript.
When to use:
.xlsm files that fail in openpyxlKey features:
openpyxlzipfile + xml.etree (no heavy dependencies)definedNames repair workflow for problematic filesExample usage:
# Install the documentation suite
claude plugin install daymade-docs@daymade-skills
# Then ask Claude to automate Excel workflows
"Create a formatted valuation template workbook"
"Parse this .xlsm and extract the DCF sheet"
"Generate an AppleScript sequence to zoom and scroll Excel before screenshot"
🎬 Live Demo
Coming soon
📚 Documentation: See excel-automation/SKILL.md and formatting-reference.md.
Requirements: Python 3.8+, uv, openpyxl (auto via uv run --with openpyxl), macOS for AppleScript window control.
Install:
claude plugin install daymade-macos@daymade-skills(suite-only — invoked asdaymade-macos:capture-screen)
Capture application windows by CGWindowID with a reliable three-step workflow: discover window IDs via Swift, control app state via AppleScript, and capture outputs with screencapture.
When to use:
Key features:
CGWindowListCopyWindowInfo)Example usage:
# Install the Apple platform suite
claude plugin install daymade-macos@daymade-skills
# Then ask Claude to capture windows programmatically
"Find the Excel window ID and capture it silently"
"Create a multi-shot capture workflow for this workbook"
"Capture Chrome window sections with scripted scrolling"
🎬 Live Demo
Coming soon
📚 Documentation: See capture-screen/SKILL.md.
Install:
claude plugin install daymade-macos@daymade-skills(suite-only — invoked asdaymade-macos:macos-permissions)
Use for repeated TCC prompts, silent denials, and background jobs that cannot read protected files. It identifies the actual requester, checks for a usable existing grant, and verifies the repair through the real job. The Skill contains the diagnostic and Full Disk Access repair procedure.
📚 Documentation: See macos-permissions/SKILL.md.
Requirements: macOS. Reading TCC.db also requires Full Disk Access for the process doing the read.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:continue-claude-code-work)
Continue a verified Claude Code Session without reopening the old interactive
session. The continuation layer first consumes read-claude-code-history, then
reconstructs the original business outcome, unfulfilled requests, corrections,
proven prior assets, and one next action that directly advances the goal.
When to use:
.claude JSONL files instead of running claude --resumeKey features:
Example usage:
# Then ask Claude to resume from local artifacts
"continue work from session 123e4567-e89b-12d3-a456-426614174000"
"don't resume, just read the .claude files and continue"
"check what I was working on in the last session and keep going"
📚 Documentation: See continue-claude-code-work/SKILL.md.
Requirements: Python 3.8+, git for workspace reconciliation.
Install, troubleshoot, and use Scrapling CLI with a verified static-first workflow for extracting HTML, Markdown, or text from webpages. Includes a diagnostic script for broken extras installs, Playwright browser runtime checks, and smoke tests against real URLs.
When to use:
uv tool install scrapling, or scrapling extractmp.weixin.qq.com)Key features:
diagnose_scrapling.py script for CLI, browser runtime, and live URL smoke testsextract get, escalate to extract fetch only when needed#js_content for clean article Markdownclick, Playwright runtime setup, and curl: (60) trust-store failuresExample usage:
# Install the skill
claude plugin install scrapling-skill@daymade-skills
# Then ask Claude to work through Scrapling for you
"Install Scrapling CLI and verify the setup"
"Extract this WeChat article into Markdown with Scrapling"
"Decide whether this page needs static or browser-backed fetching"
🎬 Live Demo
Coming soon
📚 Documentation: See scrapling-skill/SKILL.md and scrapling-skill/references/troubleshooting.md.
Requirements: Python 3.6+, uv, Scrapling CLI, and Playwright browser runtime for browser-backed fetches.
One-stop wrapper for the official Tencent IMA skill (ima.qq.com). Installs upstream ima-skill to Claude Code, Codex, and OpenClaw via npx skills add, guides API key setup, detects and repairs known upstream issues under user consent, and implements a personalized fan-out search strategy that floats priority knowledge bases to the top.
When to use:
Skipped loading skill(s) due to invalid SKILL.md warnings related to ima-skillKey features:
~/.config/ima/{client_id, api_key} with env-var fallbackscripts/diagnose.sh read-only health check (install presence, credential liveness, known issues)scripts/search_fanout.py client-side cross-KB search with priority lists, subset-skip lists, and 100-hit silent-truncation detectionMODULE.md or prepend minimal frontmatter)~/.config/ima/copilot.json with illustrative-only template valuesExample usage:
# Install the skill
claude plugin install ima-copilot@daymade-skills
# Then ask Claude to drive the flow
"Install ima-skill and configure my IMA API key"
"Run diagnose on my ima-skill and fix whatever is broken"
"Search my IMA knowledge bases for embedding model comparisons, priority to my curated KB"
🎬 Live Demo
Coming soon
📚 Documentation: See ima-copilot/SKILL.md and ima-copilot/references/known_issues.md.
Requirements: Node.js 18+ (for npx skills), curl, unzip, Python 3.6+. IMA OpenAPI credentials from https://ima.qq.com/agent-interface.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:fixing-claude-export-conversations)
Reconstruct broken line wrapping in Claude Code exported .txt conversation files. Rebuilds tables, paragraphs, paths, and tool calls that were hard-wrapped at fixed column widths, and ships with an automated 53-check validation suite (file-agnostic, catches over- and under-merging regressions).
When to use:
YYYY-MM-DD-HHMMSS-*.txt/export output before sharing or archiving itKey features:
fix-claude-export.py) with --stats mode for before/after metricsvalidate-claude-export-fix.py) that catches regressionsuv and Python 3.8+Example usage:
# Fix and show stats
uv run daymade-claude-code/claude-export-txt-better/scripts/fix-claude-export.py broken.txt --stats
# Custom output path
uv run daymade-claude-code/claude-export-txt-better/scripts/fix-claude-export.py broken.txt -o fixed.txt
# Validate the fix
uv run daymade-claude-code/claude-export-txt-better/scripts/validate-claude-export-fix.py broken.txt fixed.txt
🎬 Live Demo
Coming soon
📚 Documentation: See claude-export-txt-better/SKILL.md and the bundled evals/ fixtures.
Requirements: Python 3.8+, uv package manager.
Export and sync Douban (豆瓣) book / movie / music / game collections to local CSV files via the reverse-engineered Frodo API. Full export covers all history; RSS incremental sync keeps daily updates current. No login, no cookies, no browser — just a user ID and it works.
When to use:
Key features:
.gitleaks.toml allowlist for the public Android APK credentialsExample usage:
# Full export of user's collections
uv run douban-skill/scripts/douban-frodo-export.py <douban-user-id>
# Incremental RSS sync (last ~10 items per category)
uv run douban-skill/scripts/douban-rss-sync.py <douban-user-id>
🎬 Live Demo
Coming soon
📚 Documentation: See douban-skill/SKILL.md and douban-skill/references/troubleshooting.md for the complete failure log of rejected approaches.
Requirements: Python 3.8+, uv package manager. No login or cookies required.
Designs and diagnoses safe Terraform releases as well as the provisioner traps learned from real incidents. It keeps staging and production on one required configuration schema, validates exact candidate bytes + the Compose-rendered environment + the immutable runtime before live mutation, and binds saved plans to staging evidence, source provenance, explicit production authorization, and independent readback.
When to use:
null_resource provisioners or remote-exec blocks that SSH into fresh instancesterraform applyKey features:
Example usage:
# Trigger the skill naturally during Terraform work
"I'm getting 'docker: not found' in my null_resource provisioner after apply"
"My rsync local-exec is failing with 'connection unexpectedly closed'"
"Help me write a multi-env Terraform setup without snapshot cross-contamination"
"Staging has this Caddy variable but production leaves it empty — how do I validate both safely?"
🎬 Live Demo
Coming soon
📚 Documentation: See terraform-skill/SKILL.md and the bundled references/ for detailed remediation patterns.
Requirements: None (Terraform-adjacent knowledge only; no runtime dependencies).
Guides users through structured narrative design (ABCDEFG model), then delegates visual generation to baoyu-slide-deck. Focuses on what machines can't do — narrative co-design with humans.
When to use:
Key features:
Example usage:
# Trigger the skill naturally
"Help me turn my article into a slide deck"
"Create a presentation from my talk transcript"
"I need a 20-minute deck for a workshop"
Requirements: baoyu-slide-deck skill for visual generation.
Batch-place existing images onto an Excalidraw whiteboard, laid out on a generous grid so nobody has to drag them apart afterwards. Also turns a slide deck into clean per-slide images first, and inspects what is inside a .excalidraw file. The available Excalidraw MCP servers and skills cover element CRUD and export but document no image element type, no dataURL handling, and no files map — embedding your own pictures is the gap this fills.
When to use:
Key features:
--exclude for images already on the board--template-from copies the image-element field set out of your own board — Excalidraw's published schema stops before fileId/status/scale/cropinnerText still reads hidden fragments as presentExample usage:
# Trigger the skill naturally
"Put these screenshots on my Excalidraw board"
"Add my old workshop images to the canvas, spaced out"
"Turn this deck into images I can draw on"
Note: Not for generating a diagram from a text description — that is a different job.
Falsification-first methodology for network, streaming, and protocol-layer bugs where the obvious cause is probably wrong. Built from a real 5-hour SSE incident where assumption-stacking wasted hours that a 10-minute layered experiment would have resolved.
When to use:
ECONNRESET, HTTP/2 RST_STREAM, INTERNAL_ERROR)ERR_CONNECTION_CLOSED, SSL_ERROR_SYSCALL, fake TUN DNS IPs, CNAME-based rule overrides)UNKNOWN_CERTIFICATE_VERIFICATION_ERROR, wrong-site certificate)Key features:
layered-isolation-probe.sh, mock-idle-upstream.py)ERR_CONNECTION_CLOSEDRequirements: None (methodology + portable shell/Python probes).
Install:
claude plugin install daymade-audio@daymade-skills(suite-only — invoked asdaymade-audio:stepfun-tts)
Generate Chinese / Japanese speech with stepaudio-2.5-tts. Captures the two non-obvious TTS pitfalls that cost hours otherwise: voice_label removal (replaced by natural-language instruction) and stricter 2.5-era censorship (死/消失/political terms).
When to use:
censorship_block fallbackstep-tts-2 to stepaudio-2.5-tts (voice_label → instruction breaking change)Key features:
stepaudio-2.5-tts with instruction (≤200 chars natural-language mood) + inline () prosodytts_generate.py (with --batch <jsonl>) and ab_compare.sh$STEPFUN_API_KEY → ${CLAUDE_PLUGIN_DATA}/config.json fallbackreferences/migration_from_v2.mdRequirements: StepFun API key, "Normal" tier (https://platform.stepfun.com/). For ASR / transcription, use the sibling stepfun-asr skill below.
Install:
claude plugin install daymade-audio@daymade-skills(suite-only — invoked asdaymade-audio:stepfun-asr)
Transcribe Chinese / English audio with stepaudio-2.5-asr. Hides the #1 trap of the 2.5 ASR family: it does NOT live on /v1/audio/transcriptions — the wrong endpoint returns a misleading model stepaudio-2.5-asr not supported error that looks identical to a permission/whitelist failure.
When to use:
step-asr / step-asr-1.1 (different endpoint, different body shape, SSE response)model stepaudio-2.5-asr not supported error (= wrong endpoint, not permission)Key features:
/v1/audio/asr/sse SSE streaming with base64 audio + nested JSON body (the script handles all four traps)asr_transcribe.py — pure-stdlib CLI, auto-detects mp3/wav/ogg/opus/pcm by extensionerror events (censorship can fire on ASR side too — rare but real)$STEPFUN_API_KEY → ${CLAUDE_PLUGIN_DATA}/config.json fallbacktranscript-fixer (ASR error correction) and meeting-minutes-taker (structured minutes) as natural downstream skillsRequirements: StepFun API key, "Normal" tier (https://platform.stepfun.com/). Plan keys cannot call audio endpoints.
Make a repository runnable and handoff-ready without guessing its stack or changing how collaborators normally work. The skill reads project authority first, repairs the verified gap, and treats startup instructions, lifecycle hooks, and Git mutation as different mechanisms with different safety boundaries.
When to use:
Key features:
check_env.py infers only declared toolchains from manifests/lockfiles; it does not assume ffmpeg, uv, Python, or .envExample usage:
# Install the skill
claude plugin install auto-repo-setup@daymade-skills
# Then ask Claude naturally
"我跑不起来这个仓库"
"帮我设置一下这个项目的环境"
"进入项目先同步远端；本地有改动不要自动 stash"
"为什么同一个 SessionStart 输出了三次？先查清来源"
"只有首条消息前必须注入动态提醒时，才帮我装 hook"
"git push 被拒了"
Requirements: The guidance itself has no runtime dependency. Bundled Python utilities require Python 3.10+; no external API key is required.
Render a terminal CLI program's colored output to a PNG so Claude can actually see the rendered result — color contrast, alignment, background blocks, highlighting — instead of only reading plain text and raw ANSI escape codes. Reading a hex value is guessing; seeing the rendered contrast on the real terminal background is verification.
When to use:
Key features:
render_ansi.sh (freeze/Chrome auto-select), ansi2html.py (stdlib renderer)Example usage:
# terminal-screenshot lives in the daymade-claude-code suite
claude plugin install daymade-claude-code@daymade-skills
# Then ask Claude naturally
"verify my delta diff colors"
"看一下这个终端配色的真实效果"
"is the add/remove contrast in git diff strong enough?"
Requirements: macOS. charmbracelet/freeze (preferred renderer) or Google Chrome (fallback). Python 3 for the fallback renderer.
Convert a PDF into one self-contained, readable HTML file that preserves images, charts and reading order — optionally translating it into another language while keeping every figure. A PDF is a layout, not just a text stream, so the workflow renders each page for you to see before building, and renders the HTML for visual verification before delivery.
When to use:
Key features:
Example usage:
# pdf-to-html lives in the daymade-docs suite
claude plugin install daymade-docs@daymade-skills
# Then ask Claude naturally
"把这个 PDF 转成中文网页版"
"make this report readable as HTML"
"translate this PDF to English but keep the charts"
Requirements: uv, Google Chrome or Chromium (visual verification). Python packages (PyMuPDF, Pillow, numpy) auto-install via uv run --with.
Install:
claude plugin install daymade-audio@daymade-skills(suite-only — invoked asdaymade-audio:asr-transcribe-to-text)
Transcribe audio and video files to text using Qwen3-ASR via two interchangeable inference paths: local MLX on macOS Apple Silicon (no API key, 15-27x realtime) or a remote vLLM/OpenAI-compatible API for any platform. Auto-detects the platform and recommends the best path, persisting the choice in ${CLAUDE_PLUGIN_DATA}/config.json.
When to use:
Key features:
transcribe_local_mlx.py loads the model once and processes files sequentially (no GPU contention)overlap_merge_transcribe.py splits into 18-minute chunks with 2-minute overlap and fuzzy-mergestranscript-fixer to clean ASR recognition errors on the outputExample usage:
# asr-transcribe-to-text lives in the daymade-audio suite
claude plugin install daymade-audio@daymade-skills
# Then ask Claude naturally
"transcribe this meeting recording to text"
"把这个录音转成文字"
"convert lecture.mp4 to a transcript"
Requirements: uv, ffmpeg/ffprobe. Local MLX path needs macOS Apple Silicon; remote path needs a reachable vLLM/OpenAI-compatible ASR endpoint. No API key for local mode.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:marketplace-dev)
Create and maintain Claude Code plugin marketplaces: convert a repository, consolidate standalone skills into a new or existing suite, move skills between suites, validate real installation/cache boundaries, and ship the result through a PR.
When to use:
claude plugin installmarketplace.json (plugin distribution, one-click install, auto-update)Unrecognized key: "$schema" or duplicate plugin namesKey features:
$schema is rejected, metadata has only 3 valid fields, strict: false semantics, single-skill vs suite source/skills patternscheck_marketplace.sh runs four checks (JSON syntax → claude plugin validate → source/skills resolution → reverse sync) and exits non-zero on failuresource produced the intended snapshotmarketplace.json edit; warn on un-bumped version when a SKILL.md changes) that auto-activate with the pluginExample usage:
# marketplace-dev lives in the daymade-claude-code suite
claude plugin install daymade-claude-code@daymade-skills
# Then ask Claude naturally
"turn this skills repo into a plugin marketplace"
"generate a marketplace.json for this repo and validate it"
"add my new skill to the marketplace and open a PR"
"move these standalone skills into daymade-macos and make them suite-only"
Requirements: claude CLI (for claude plugin validate / install tests), jq. Git remotes configured if opening an upstream PR.
Install:
claude plugin install daymade-skill@daymade-skills(suite-only — invoked asdaymade-skill:skill-creator)
The essential meta-skill for building your own skills. It scales verification to the change: bounded fixes get targeted checks, narrow behavior changes get sampled replays, and broad/high-risk work becomes eligible for heavier evidence without automatically starting it. Paired baselines, graders, benchmarks, and viewers require explicit authorization or a decision-bearing evidence plan plus opt-in. It also supports explicit benchmarking and optimizes a skill's description for better triggering accuracy.
When to use:
Key features:
context: fork decision guide (subagents can't spawn subagents or call skills) and composable/orthogonal skill designinit_skill.py scaffolding, package_skill.py (auto-validates), and security_scan.py (gitleaks-based secret/PII detection)Example usage:
# skill-creator lives in the daymade-skill suite
claude plugin install daymade-skill@daymade-skills
# Then ask Claude naturally
"create a skill that does X"
"improve this skill's description so it triggers more reliably"
"benchmark this skill against a no-skill baseline"
Requirements: Python 3, uv, PyYAML (validation/packaging), gitleaks (security scan). claude CLI only for agent evals and description-optimization runs.
Extract Feishu (Lark) Docs, Wiki pages/collections, spreadsheets (including cell-attachment file download), and Minutes (妙记) transcripts into faithful local Markdown. The primary path uses the lark-cli API — it extracts the document body programmatically (no model paraphrasing), recursively follows a collection's reference graph, and reads permission boundaries from error codes; a browser-DOM path is the fallback only when lark-cli cannot reach the content.
When to use:
.docx into faithful Markdown with heading/highlight restorationKey features:
jq (never retyped by the model — the single most important fidelity rule)feishu_extract_refs.py, plus a residual rich-media-tag acceptance gate so no referenced doc is silently missed.docx → Markdown with font-size→heading and w:shd→highlight restoration, then visual verificationLARK_CLI_NO_PROXY=1 discipline for *.feishu.cn (avoids credential leak/DNS hijack) and a U+FFFD encoding-corruption final checkExample usage:
# Install the skill
claude plugin install feishu-doc-scraper@daymade-skills
# Then ask Claude naturally
"把这个飞书合集导出成 markdown"
"export this Feishu Minutes transcript"
"save this Lark wiki page as Markdown"
Requirements: lark-cli binary (npm @larksuite/cli) authenticated to the target tenant; jq. Fallback path needs a browser-automation surface; the docx path needs python-docx and a docx→md converter (the bundled doc-to-markdown skill or pandoc).
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:financial-router)
Routes Bigdata/RavenPack, free US company fundamentals, A-share news, sector Top N and announcements, and Gangtise setup to one bundled specialist. Pharmaceutical reporting and general argument review keep their direct Skills; all specialist slash commands remain available for manual use.
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:bigdata-skill)
Pull Bigdata.com (RavenPack) financial and news data through the official bigdata-client SDK and its public /v1/* REST endpoints — reaching the structured substrate the Bigdata MCP server doesn't hand over. The MCP returns prose chunks and pre-synthesized tearsheets; this toolkit reaches structured financials, prices, analyst estimates, a daily entity-sentiment series, annotated chunk search with sentiment + entity spans, and a screener.
When to use:
bd_v2_ API key, rp_entity_id, query_unit/chunk cost, bigdata-client, or "the bigdata MCP isn't enough"Key features:
BigdataClient exposing both the SDK (search + knowledge graph) and a REST escape hatch (bd._api.http) for every /v1/* endpoint the SDK never wrappedfields_values_to_records() to flatten {fields, values} responses1 query_unit = 10 chunks, only chunk-search billed, ChunkLimit (never a bare int), rerank thresholds, 50%-cheaper batch search, and a CostModel/CostTracker budget vetorc() SSL-retry wrapper for the common first-handshake SSL: UNEXPECTED_EOF, plus a known-pitfalls reference with reproductions and fixesBIGDATA_API_KEY (no plaintext fallback); read-only, never writes/uploadsExample usage:
# Install the suite
claude plugin install daymade-financial@daymade-skills
export BIGDATA_API_KEY=bd_v2_xxxxxxxx
# Then ask Claude naturally
"pull NVIDIA's forward analyst estimates and last earnings surprise from Bigdata"
"give me a daily entity-sentiment series for this ticker"
"the bigdata MCP only gave me a tearsheet — I need the structured fields"
Requirements: A bd_v2_ Bigdata.com API key (env var, never hardcoded), uv, the official bigdata-client SDK in an isolated venv. Optional outbound/WSS proxy only if your network needs one to reach api.bigdata.com.
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:gangtise-copilot)
One-command installer, credential configurator, and diagnostic layer for the full Gangtise (岗底斯投研) OpenAPI skill suite. Installs the official Gangtise skills (data, research, utility), configures accessKey/secretAccessKey with a live auth check, and runs a read-only health diagnostic — solving the suite's core discoverability problem (no public manifest, listing-disabled OBS bucket, parallel naming lines).
When to use:
gangtise-* skilltoken is invalid / 接口地址错误, or "my gangtise install is broken"Key features:
install_gangtise.sh downloads the declared OBS bundles → extracts skill directories → symlinks them into detected agent skills dirs (Claude Code, OpenClaw, Codex), with minimal/workshop/full/--only presetsconfigure_auth.sh writes a shared XDG credential file (mode 600), runs a live auth call, and symlinks every skill's .authorization to itdiagnose.sh reports install state, credential validity, and scoped capability tiers (auth scope vs RAG scope)Example usage:
# Install the suite
claude plugin install daymade-financial@daymade-skills
# Then ask Claude naturally
"装一下 gangtise 的所有 skill 并配置好凭据"
"my gangtise skills report token is invalid — diagnose it"
"宁德时代的研报用哪个 gangtise skill 查"
Requirements: A Gangtise accessKey + secretAccessKey; bash, curl, network access to the official OBS bucket and open.gangtise.com. Works with Claude Code, OpenClaw, and Codex agent layouts.
Co-create a personal investment-research LLM Wiki (Andrej Karpathy's pattern) where the user's OWN analysis framework becomes a living CLAUDE.md — built by interviewing them rather than handing over a template. Pure markdown + [[wikilinks]]; knowledge is compiled into pages instead of being re-retrieved from raw files on every query (Karpathy's core idea — no RAG pipeline in place of compilation; a local search engine such as qmd over the grown wiki is part of his pattern). The value is extracting the user's personal investment preferences into THEIR OWN schema, never imposing a standard one.
When to use:
Key features:
init_vault.py scaffolds the mechanism layer only (no schema), then an 8-dimension interview builds the user's own CLAUDE.md in their own wordslint-vault.py keep the vault consistent and fight derived-value driftanalyst-track-record skill and Bash) and chains into analyst-track-record for analyst back-testing — without rebuilding itExample usage:
# Install the skill
claude plugin install llm-wiki-setup@daymade-skills
# Then ask Claude naturally
"帮我搭一个投研第二大脑"
"build me a personal investment-research wiki in Karpathy's style"
"ingest this earnings call into my research vault"
Requirements: Python 3, uv (for init_vault.py / lint), git. Markdown + wikilinks; no vector DB or embedding service required (qmd is an optional local search layer once the wiki grows). Pairs with the analyst-track-record skill for back-testing.
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:benchmark-due-diligence)
Run adversarial due-diligence on a benchmark the user envies — a founder, KOL, company, or product whose claimed success looks inflated — separating marketing bubble from real signal, then mapping the validated playbook onto the user's own resources. The adversarial, decision-oriented cousin of deep-research: it assumes the picture is inflated until proven otherwise and ends in "what this means for ME", not a neutral report.
When to use:
deep-research's neutral briefingKey features:
坐实/存疑/证伪-水分 verdicts) → due-diligence conclusion (bubble-busting table + attribution breakdown) → commissioner resource-mappingdeep-research fan-out, osint-investigate identity checks, the qcc family for 工商 data, agent-reach for social-platform data)context: fork would silently break the fan-out)Example usage:
# Install the financial research suite
claude plugin install daymade-financial@daymade-skills
# Then ask Claude naturally
"帮我尽调一下这个创始人，他到底有没有水分"
"tear down this competitor's playbook and tell me what I can actually copy"
"this KOL claims 0-to-1M users — is that real, and is it replicable for me?"
Requirements: Web access for the collection/verification agents. Optionally composes with deep-research, osint-investigate, the qcc skill family, and agent-reach; renders a shareable report via pdf-creator.
Fetch real, citable data for any Bilibili (B站) video — title, UP follower count, publish date, tags, partition, per-part cids, live stats (view/like/coin/favorite/share/reply/danmaku), and full danmaku (bullet-comment) text — in one view/detail call, login-free. Built so engagement numbers are cheap to fetch and impossible to fake, instead of hand-typed into a doc where they rot.
When to use:
av number, b23.tv short link, or full URL — all normalized automaticallyKey features:
bili-fetch.sh returns full metadata + live stats + UP fans + tags + every part's cid; metrics carry a fetched_at timestamp because they drift in real timebili-danmaku.sh pulls and decompresses the danmaku full text; bili-subs.sh handles the login-gated subtitle track (asks before touching browser cookies)bili-selftest.sh health-check verifies every endpoint against the live API, so API drift surfaces as one clear FAIL instead of a silent wrong answerx/v3/fav/*, verified no-WBI), the verified SESSDATA subtitle path, and the WBI request-signing algorithm for space/wbi/* extensionExample usage:
# Install the skill
claude plugin install bilibili-source@daymade-skills
# Then ask Claude naturally
"pull the real view/like/favorite counts for this B站 video so I can cite them"
"这个 B站 视频弹幕里大家在说什么？"
"grab the subtitle transcript from this bilibili video so I can summarize it"
Requirements: curl, jq, python3 (danmaku decompression). yt-dlp only for the login-gated subtitle path. No login for stats / metadata / danmaku.
Preserve and recover recorded Ghostty CLI sessions through the canonical Skill. For scheduled metadata backups, partial coverage and operating prerequisites, use the automatic workflow.
claude plugin install ghostty-use@daymade-skills
# Then ask Claude naturally
"我要重启电脑更新，把终端会话都保存一下"
"恢复之前的 ghostty 窗口和会话"
"restore my ghostty tabs after the update"
"检查一下有没有会话丢了"
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:claude-usage-analyst)
Turn local ccusage data into an evidence-based, human-readable explanation of where your Claude Code / Claude Desktop tokens, cost, and quota went — separating observed numbers from interpretation instead of guessing.
When to use:
fable / opus / sonnet) is unusually expensive for your workloadKey features:
analyze_claude_usage.py summarizes tokens, cost, input/output, and cache create/read over any date window and timezone--model-a / --model-b) weighs both token volume and estimated cost — a model can be cheap per token but expensive overallccusage output; cache-read pressure is counted even when you never typed those tokensccusage claude measures local Claude Code logs, not a full Claude.ai chat billExample usage:
# Install the suite
claude plugin install daymade-claude-code@daymade-skills
# Then ask Claude naturally
"why did my Claude quota run out today?"
"is opus more expensive than sonnet for what I'm doing?"
"break down my Claude Code token usage for this month"
Requirements: ccusage (via npm i -g ccusage or npx ccusage@latest), python3.
claude plugin install marketplace-health-check@daymade-skills
Run a comprehensive, evidence-based health check of this skills marketplace repo with a parallel fan-out Dynamic Workflow — six inspectors cover code/script safety, documentation/SSOT consistency, security/PII leaks, open-PR triage, open-issue triage, and marketplace-manifest integrity at once — then the serious findings are Counter-Reviewed before they reach the report.
When to use:
Key features:
Example usage:
# Install
claude plugin install marketplace-health-check@daymade-skills
# Then ask Claude naturally
"do a full health check of this repo before I cut a release"
"audit the marketplace — code, docs, PII, PRs, issues, everything"
"全面体检一下这个仓库"
Requirements: gh CLI (authenticated), git, jq, python3; opt-in to the Workflow tool (asking to run the health check is the opt-in).
claude plugin install daymade-claude-code@daymade-skills
Set up multiple isolated Claude Code CLI profiles so you can run different LLM providers (Kimi, MiniMax, GLM, DeepSeek, StepFun, Anthropic) in separate terminal windows at the same time — each profile gets its own .claude.json state while sharing skills, projects, hooks, and agents.
When to use:
Key features:
~/.config/claude-switch-models-setup/ and seeds an empty Codex activation manifest only when none exists~/.claude/settings/<provider>.json templates with required isolation flagsclaude-profiles-init creates isolated ~/.claude-profiles/<provider>/ directories with symlinked shared resources~/.agents/skills; a declared active subset may keep legacy ~/.codex/skills compatibility linksExample usage:
# Install the suite
claude plugin install daymade-claude-code@daymade-skills
# Then ask Claude naturally
"set up Claude Code profiles for Kimi and DeepSeek"
"I want to run Kimi and Anthropic in separate terminals"
"install the multi-provider profile setup from the workshop"
Requirements: claude CLI, zsh or bash, python3, plus API keys for the providers you want to use.
claude plugin install llm-eval-harness@daymade-skills
Evaluate any LLM behind an OpenAI- or Anthropic-compatible endpoint across four dimensions — instead of trusting a vendor's headline numbers — and report measured results separately from inferred ones.
When to use:
Key features:
speed_probe.py — TTFT + thinking-aware tokens/sec), concurrency/stability (concurrency_probe.py — success rate, p50/p90, breaking point), Anthropic protocol compliance (protocol_probe.py — thinking-block trigger rate over N≥10), and quality regression (usecase_runner.py + independent blind judges)reasoning_content separately so reasoning tokens don't inflate tok/s (the trap that once read a ~750 tok/s model as 4700)fully-implemented / intermittent (k/N) / not-implemented, never concluding from a single sample, with Connection: close so a load balancer can't pin all samples to one replica--key-env MY_KEY) — the key never appears in ps, shell history, or a saved report~/.llm-eval/usecases.json) so it survives skill updates and never lands in a public repoExample usage:
export MY_KEY=sk-... # the key never appears in a command below
# Speed: real-task throughput + sustained decode ceiling
uv run --with openai python scripts/speed_probe.py \
--base-url https://api.example.com/v1 --model some-model --key-env MY_KEY --mode both
# Concurrency: ramp until it breaks
uv run --with aiohttp python scripts/concurrency_probe.py \
--url https://api.example.com/v1/chat/completions --model some-model \
--key-env MY_KEY --format openai --concurrency 10 20 40 60
🎬 Live Demo
Coming soon
📚 Documentation: See llm-eval-harness/references/evaluation_disciplines.md for the reasoning behind each discipline and llm-eval-harness/references/quality_blind_judge.md for the blind-judge method.
Requirements: Python 3.8+, uv; openai and aiohttp (auto-installed via uv run --with); an API key for the endpoint under test. Optionally composes with promptfoo-evaluation for rubric-based gating.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:read-claude-web-conversation)
Extract the complete active Claude.ai conversation through the Claude Code operations suite: message text, real tool calls and outputs, uploads, generated images, and sandbox deliverables. Render faithful Markdown locally, reconstruct files created through tool calls, or download original bytes through the logged-in browser session.
When to use:
Requirements: Installed daymade-claude-code suite and a logged-in Chrome
session. The AppleScript fallback requires macOS and one user-enabled Chrome
developer-menu toggle.
claude plugin install setup-notifications-via-wecom@daymade-skills
Set up reusable WeCom (Enterprise WeChat) webhook notifications for technical status reports, alerts, and completion messages. The target is explicitly classified: the user's own channel may send automatically; every other target requires human confirmation.
For recurring reports on an existing SLS pipeline, follow the bundled adapter guide for truncation limits, evaluation dates, recipient conflicts and native delivery acceptance.
When to use:
Requirements: WeCom bot webhook URL and shell access.
claude plugin install notify-wecom@daymade-skills
Send a single WeCom group-bot message using the configured target identity: self sends directly, while others requires human confirmation.
When to use:
/notify-wecomRequirements: WeCom bot webhook URL plus the explicit recipient scope/label and canonical sender binding created by the setup command.
claude plugin install github-sensitive-data-cleanup@daymade-skills
Scan and remove sensitive data from GitHub repository history, with backup, visibility checks, and force-push safety gates.
When to use:
Requirements: git, GitHub access, and the relevant scanning/history-rewrite tools for the target repository.
Install:
claude plugin install daymade-codex@daymade-skills(suite-only — invoked asdaymade-codex:codex-image-gallery)
Start a self-contained local web gallery for Codex-generated image outputs. The skill bundles its Node server and HTML UI, scans ~/.codex/generated_images by default, and can point at another folder with GALLERY_ROOT.
When to use:
~/.codex/generated_imagesKey features:
scripts/server.mjs and assets/index.html/api/images scan; no hardcoded manifest/images/<relative-path> image route with path traversal protectionGALLERY_ROOT, PORT, and HOSTRequirements: Node.js 18+ and access to the image folder.
claude plugin install frontend-visual-qa@daymade-skills
Audit the UI users can actually see, with explicit evidence levels and no source mutation by default.
When to use:
ui-designer/design work and the broader process managed by qa-expertKey features:
claude plugin install openclaw@daymade-skills
Manage OpenClaw (龙虾/lobster) instance configurations — audit, diff, copy, add-model, list, and switch models across openclaw.json files.
When to use:
Key features:
--no-audit escape hatchclaude plugin install download-gemini-images@daymade-skills
Download images (uploaded files or generated previews) from a Google Gemini conversation page using your logged-in Chrome session, then package them into an ordered ZIP.
When to use:
Key features:
pageAssets fallback when lightbox automation failsclaude plugin install wps-doc-scraper@daymade-skills
Faithfully archive public WPS / KDocs / 金山文档 links — especially embedded ProcessOn mind maps and canvases — as raw source data, original SVG/PNG, and Markdown, without logging in.
When to use:
kdocs.cn or wps.processon.com link to scrape, save, download, or convert to MarkdownKey features:
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:ashare-news-fetcher)
Aggregate A-share (Chinese stock market) news, policy, and sentiment from public sources — 财联社, 华尔街见闻, 金十, 新浪 7x24, 东财快讯, regulator announcements (CSRC / PBoC / SSE), 东方财富股吧 — into structured JSON or Markdown.
When to use:
Key features:
Install:
claude plugin install daymade-codex@daymade-skills(suite-only — invoked asdaymade-codex:local-codex)
Delegate coding tasks to the local OpenAI Codex CLI agent using your ChatGPT Pro OAuth flat-rate subscription. Wraps codex exec / codex review for code generation, refactoring, and review without per-token API charges.
When to use:
codex exec for non-interactive coding tasksKey features:
~/.codex/auth.json (no API keys)claude plugin install openclaw-model-switch@daymade-skills
Switch the default AI model for an OpenClaw instance (e.g., Kimi K2.6 → K2.7) by safely editing openclaw.json with automatic backup, model validation, and optional gateway restart.
When to use:
Key features:
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:pharma-daily-report)
Generate an A-share pharmaceutical sector daily report — pull real-time quotes for core pharma stocks from Sina Finance, rank 7 sub-sectors, top gainers/losers, and estimated fund flow, then optionally push a rich-text report via Feishu.
When to use:
Key features:
Install:
claude plugin install gemini-history-analyzer@daymade-skills
Analyze Google Takeout exports of Gemini conversation history — extract and categorize transcripts and attachments, mine them for domain-specific insights with context-verified keyword search, and optionally distill findings into a personal knowledge base.
When to use:
Key features:
unar, not the corrupting macOS unzip)Install:
claude plugin install daymade-skill@daymade-skills(suite-only — invoked asdaymade-skill:skill-governance)
Govern the Skill surface Claude Code and Codex actually expose without losing cold capability. It separates canonical source, installed inventory, discovery policy, the fresh model-visible catalog, and router-resolved runtime resources.
When to use:
~/.agents/skills without treating every disk bundle as activeKey features:
skills/list, detecting truncation, omitted enabled entries, identity/source collisions, disabled-path drift, scan errors, and missing direct activation entriesInstall:
claude plugin install daymade-docs@daymade-skills(suite-only; invoke asdaymade-docs:photo-to-scanned-pdf)
Turn phone photos of contracts, receipts, forms, certificates, or handwritten pages into a clean A4 PDF. The workflow combines perspective correction, scanner-style background cleanup, colored-paper handling, explicit content-based page ordering, and a mandatory whole-document contact-sheet check.
claude plugin install github-review-pr@daymade-skills
Review or re-review one named contributor PR—including a closed PR under reconsideration—or sweep every open PR newest-to-oldest as a repository maintainer, always against the live base rather than a stale PR snapshot.
When to use:
Key features:
PR, BASE, or SHARED before assigning repair ownershipDECLINE from a fix request or true supersessionExample usage:
/github-review-pr --personal-maintainer https://github.com/<owner>/<repo>/pull/<number>
/github-review-pr --personal-maintainer --all-open
main changed since the last review; tell me what would actually land now
review all of our open PRs newest to oldest and apply my prior maintainer principles
can we merge this contribution and fix the remaining repo bookkeeping ourselves?
Requirements: authenticated gh CLI, git with merge-tree --write-tree, and jq.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:read-codex-history)
Read, search, and export local Codex history without continuing the old task. It keeps the prompt ledger, state database, and rollout JSONL as three distinct evidence surfaces so exact user input, inventory metadata, and Agent behavior are never silently substituted for one another.
Key features:
session_meta.id equality and reports fused/missing rollout gaps instead of guessingExample usage:
/daymade-claude-code:read-codex-history
list the recent Codex chats for this folder
show my 200 most recent original Codex inputs, grouped by Session
read Codex Session 01abc... chronologically and show its fork lineage
show which Codex sessions in this workspace have held canonical writer-lock files
show Codex threads including archived conversations as JSON
📚 Documentation: See storage_and_portability.md for source selection, path normalization, privacy boundaries, and diagnostics.
Requirements: Python 3.10+; no third-party packages or network access.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:continue-codex-work)
Continue a prior Codex task without replaying it through codex resume.
The Skill first requires read-codex-history to prove selected identity, fork
lineage, chronology, compacted context, and gaps; it then rebuilds the original
business outcome, unfulfilled requests, user corrections, proven assets, and one
direct next action before changing the project.
It is for a new/different Agent context taking over an earlier rollout. When
Codex itself natively resumes the same conversation and prior context is already
present, continue directly instead of invoking this Skill.
/daymade-claude-code:continue-codex-work 019f66...
Codex was cut off mid-task; recover that rollout and finish the work
📚 Documentation: See continue-codex-work/SKILL.md.
Requirements: Python 3.10+ and local Codex rollout files.
Install:
claude plugin install git-safety-net@daymade-skills
Prevent and recover from the branch/stash/rebase tangles that strand or lose local commits, detached or dirty worktrees, and answer "is everything actually merged?" with content-level proof instead of misleading commit counts. Every command is non-destructive or additive until a step is explicitly labeled destructive.
Key features:
git reflog and git fsck (the ~90-day window)refs/dangling-backup/ before any cleanupExample usage:
/git-safety-net
did I lose any commits after all that branch switching?
is everything merged into main, or is something still stranded?
recover the commit I lost after a bad rebase
prove this worktree can be deleted without losing anything
📚 Documentation: See recovery_playbook.md, merge_verification.md, and prevention_practices.md.
Requirements: Git and a standard Bash shell. No third-party packages; current remote status requires network access, while offline audits use cached remote-tracking refs.
Install:
claude plugin install daymade-codex@daymade-skills(suite-only — invoked asdaymade-codex:design-style-picker)
Turn vague taste into concrete visual choices. Instead of guessing one final design, batch-generate a structured set of visual directions so the user picks the style they actually want — the goal is exposing the taste boundary quickly, not reading minds.
Key features:
Example usage:
/daymade-codex:design-style-picker
I can't describe the style I want — show me a batch of options
this draft is too colorful and too generic, generate more directions
evolve our current design system without throwing away existing assets
📚 Documentation: See selection-playbook.md.
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:claude-migrate-memory-to-doc)
Migrate Claude Code personal memory (the per-project memory/ directory) into
tool-agnostic reference docs, so other AI CLIs that auto-load AGENTS.md (Codex
primarily; the architecture transfers to Cursor and others) working in the same
directory read the same user profile and collaboration preferences instead of being
blind to them.
Key features:
references/ + CLAUDE.md-inline + AGENTS.md-symlink, designed around
what each tool actually auto-loads (plain-text pointers are on-demand in both tools)codex verification, and memory cleanupcodex directlyExample usage:
/daymade-claude-code:claude-migrate-memory-to-doc
migrate my memory — Codex doesn't know who I am
my memory is locked to Claude Code, make it tool-agnostic
memory has grown bloated with content that should live in docs
turn off auto memory for this project and move everything into docs
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:docx-creator)
A thin incremental layer over minimax-skills:minimax-docx — not a document
engine itself. Adds the layer the underlying OpenXML engine doesn't ship: a
verified markdown-to-docx generator for Chinese formal documents, the
alignment-layering rule that stops justified text from stretching 甲方/乙方
info blocks and signature blocks into garbage, per-list numbering restart,
CJK font dual-slot setup, and a mandatory LibreOffice-to-PDF-to-PNG visual
verification chain (qlmanage thumbnails are banned — they hide exactly the
bugs that matter).
Key features:
minimax-docx; this skill owns Chinese formal-document typography rules and the CLI-vs-C# routing decisionscripts/Program.cs) for contracts, agreements, and 公文 with 甲乙方 info blocks, numbered clauses, signature blocks, and tablesExample usage:
/daymade-docs:docx-creator
生成 Word 文档
写合同 docx
把 markdown 转成 Word
give me a Word file for this labor contract
Install:
claude plugin install daymade-docs@daymade-skills(suite-only — invoked asdaymade-docs:read-docx-review)
Read a reviewed docx (Word/WPS) and extract every comment and tracked change into a per-item adjudication table (markdown or JSON): who commented, on which paragraph, what they said, plus revision-mode insertions and deletions. Revision-aware engine on the OpenXML SDK, so content inserted via track changes (w:ins) and paragraph-level deletions (w:del) that a bare python-docx read silently misses are captured. Read-only — never mutates the reviewer's file.
Key features:
Example usage:
提取这份 docx 的批注
对方批注完的合同回来了，读一下审阅意见
读一下修订，谁批了什么、批在哪
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:claude-code-hooks)
How to write, test, register, and debug Claude Code hooks — PreToolUse / PostToolUse / SessionStart / Stop Bash guards that enforce a rule the model would otherwise talk itself past. A prose rule in CLAUDE.md is a suggestion the completion drive can override; a hook is a wall.
Key features:
bash -n + real-JSON end-to-end testing before registering, version-controlled source with installer-owned recovery, and per-profile registration convergence with human-confirmation release gatespython3 -c blockscripts/test_hook.sh)Example usage:
/daymade-claude-code:claude-code-hooks
make a hook that blocks git push --force
turn this rule I keep forgetting into a hard gate
my hook is false-blocking a healthy command
debug this hook that's poisoning my session
Install:
claude plugin install daymade-macos@daymade-skills(suite-only — invoked asdaymade-macos:macos-watchdog)
Design, deploy, and discipline macOS launchd watchdogs — LaunchAgents / LaunchDaemons that detect a recurring problem and auto-remediate it, without the watchdog itself becoming the disturbance. Distilled from 15 production watchdogs and their incident history.
Key features:
macos-permissions, and stop semantics (unload is deprecated and gets resurrected by RunAtLoad — bootout/bootstrap/disable only)watchdog-cooldown.sh (source-able escalating cooldown + manual pause state machine), new-launchagent.sh (idempotent installer with validation), and an annotated plist templateExample usage:
make a launchd watchdog that reconnects my VPN when it drops
my watchdog keeps re-launching an app I quit — make it stop
audit my existing LaunchAgents for noise and crash loops
Install:
claude plugin install daymade-macos@daymade-skills(suite-only — invoked asdaymade-macos:macos-load-doctor)
Find what is making the Mac slow or hot, attribute it to an owner, and act within the shared-machine boundary. Load average first, then a three-reading process census, parent-chain attribution, and shape classification — the skill's job is a correct, evidence-backed attribution, not process cleanup.
Key features:
vm.loadavg before any individual debuggingload_census.sh) — children-per-parent (the leak that no orphan reaper will ever clean), cumulative CPU time (the busy loop %CPU dilutes), instantaneous %CPU (the active storm)Example usage:
everything on this Mac is slow and every agent's hooks are timing out
the machine is burning hot and I can't tell why
hundreds of node processes — where are they coming from?
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:devils-advocate)
Structured devil's-advocate pressure-testing of an investment thesis against user-supplied evidence materials — evidence-anchored dissent, not role-played contrarianism (every counterpoint must cite, verbatim).
Key features:
Example usage:
/daymade-financial:devils-advocate
stress-test my bull case on <stock> against these three research reports
what would have to be true for this thesis to be wrong?
Install:
claude plugin install daymade-financial@daymade-skills(suite-only — invoked asdaymade-financial:daymade-sector-research)
A-share (Chinese market) sector research workflow: Top-N gainers across all sector constituents, announcement-window retrieval, and market-sentiment judgment with graded evidence — executed by an Agent Team with fresh-context adversarial verification. All public, no-login data sources.
Key features:
Example usage:
/daymade-financial:daymade-sector-research
医药行业今天的 Top 10 标的
这些标的最近一个月发过哪些公告
判断医药板块现在的市场情绪
Drive the logged-in Kimi desktop app through computer-use to query its built-in company, financial, market, academic, and legal data plugins without separate API keys. Results stay source-labeled and must be independently checked before they become load-bearing data. Its separate Chat Deep Research route exports a full report and attachments.
Key features:
Example usage:
/kimi-use
用 Kimi 查这家公司的股东和最新财务数据，并逐项标明来源
操作 Kimi 客户端，用已安装插件核对这组市场数据
Install:
claude plugin install tibo-reset-codex@daymade-skills
查询重置公告，核实多个 Pro 账号的剩余额度与备用重置；有时间依据时预测下一轮，记录预测、核验与撤回，供后续判断调整。
ChatGPT 什么时候重置额度
几个 Pro 账号都用完了吗，哪个还有额度
banked reset 到了吗
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:prior-work-retrieval)
When the user explicitly asks to reuse or find earlier work, verify successful code, decisions, Skills/SOPs, meetings, WeChat archives, documents, and conversation history before producing. References to the current tests, README, files, implementation, or behavior do not trigger retrieval. Every real run leaves an auditable reuse/adapt/reject receipt; zero ranked hits never become an absence claim.
Key features:
Example usage:
/daymade-claude-code:prior-work-retrieval
we solved this before; find the code and successful path before changing anything
don't rebuild the workflow until you have checked our existing Skills and history
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:lark-cli-router)
Route Feishu/Lark/Doubao requests to the matching guide embedded in the installed
lark-cli, instead of loading every Lark domain Skill into the model catalog.
Tencent IMA stays separate under ima-skill.
Key features:
lark-cli skills listlark-cli skills readExample usage:
读取这个飞书文档
查一下这条妙记的逐字稿
lark-cli says the user identity is missing a scope
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:claude-code-ping-start-5h-quota)
Schedule a one-shot local timer (macOS) that sends one minimal claude -p message
right after your subscription quota resets, so the next 5-hour usage window starts
counting while you sleep instead of when you wake up. Dual redundancy: the detached
ping survives closing the session, and a live session is re-woken to verify the
ok receipt.
Key features:
caffeinate -is holds off idle sleep during the wait; independent ps read-back verifies the timer chainpkill cancel command (live-verified) — never kill by sleep secondsExample usage:
还有 1 小时 20 分钟重置额度，帮我定时戳一下 Claude
我去睡觉了，额度重置后帮我 ping 一下开启新窗口
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:agent-web-search-setup)
A reseller or relay that proxies Claude or Codex to another cloud (Vertex AI,
Bedrock, or an OpenAI-compatible layer) cannot execute Anthropic's server-side
web_search and web_fetch. The request still succeeds, so nothing errors — the
model simply gets no results and reports that recent things do not exist, which
reads as "the model got stupid" rather than as a broken tool. This skill probes
the endpoint to establish which built-in tools are actually dead, removes those
from the client so the model stops reaching for them, installs a client-executed
replacement it can really call, and proves it with a live query. It configures
the agent; it is not itself a search engine.
Key features:
diagnose.py (standard library only, runs before anything is installed) reads the shape of the reply rather than its text: server_tool_use means the tool really ran, a plain tool_use means it was handed back and never executed, HTTP 400 means the translation layer cannot express the tool at all--api openai asks the same question of the OpenAI Responses API, where a working endpoint returns a web_search_call item, and --api both handles a relay serving both shapesinconclusive, because every one of them would otherwise have the caller delete a working toolpermissions.allow in a project's shared .claude/settings.json is dropped in any workspace whose trust dialog has not been accepted, and the client says so on stderr alone — so an agent reading --output-format json sees a refusal with no reason, and the model tells the user it is waiting for permissiontoolu_vrtx_ Vertex AI, toolu_bdrk_ Bedrock), which is why switching models on the same endpoint does not helpWebSearch still on the menu, a model made 21 calls to it and 0 to an installed alternative whose description said the built-in was brokenExample usage:
搜索返回空的，模型说这东西不存在
我用的是中转站的 key，联网搜索好像坏了
give this agent internet access — web search does nothing
Install:
claude plugin install daymade-claude-code@daymade-skills(suite-only — invoked asdaymade-claude-code:tech-selection)
A checklist for choosing between technologies — library, framework, storage, data format, model, build-vs-buy, architecture. Runs before committing, not after. The criteria are filters, not sorters: they kill candidates that violate a principle, and survivors are decided by business-result anchoring, never by ranking. When two or more candidates survive, the output is candidates
Key features:
unknown is not pass — a candidate carrying an unverified axis does not enter the survivor setExample usage:
用哪个框架？React 还是 Vue？
这个爬虫用 requests 还是 playwright
数据存哪里？SQLite 还是 Postgres？
要不要自建，还是用现成的？
别闭门造车，先看看有没有现成的
which library should we use for this?
build or buy — evaluate the options
Install:
claude plugin install daymade-codex@daymade-skills(suite-only — invoked asdaymade-codex:codex-1m-context-window-setup)
Configure the shared Codex CLI/Desktop base config for the largest context window the selected model actually declares, up to a one-million-token request. The Skill discovers the live model contract, explains the familiar ~258K default, applies a 60% auto-compaction threshold, preserves unrelated TOML, backs up changed bytes, and rolls back if Codex strict-config validation fails.
Key features:
doctor, transactional apply, and drift-detecting verify modesmodel_context_window and model_auto_compact_token_limitExample usage:
/daymade-codex:codex-1m-context-window-setup doctor
configure Codex CLI and Desktop for the largest verified context window
verify that my long-context setup still matches the selected model
Install:
claude plugin install daymade-codex@daymade-skills(suite-only — invoked asdaymade-codex:interaction-design-board)
Generate several genuinely different HTML interaction architectures for one bounded product surface while holding business facts and the current design language constant. The bundled builder combines the self-contained candidates into a keyboard-accessible Design Board with Focus/Compare modes, structured task feedback, remix requests, and a content-bound approval handoff. Product code stays untouched until the user has operated and approved a direction.
Key features:
Example usage:
/daymade-codex:interaction-design-board
show me three clickable ways to organize this review workflow before changing the app
the visual style is fine, but compare queue-first, object-first, and ledger-first interactions
use a Design Board so I can try each progressive-disclosure option and request a remix
📚 Documentation: See interaction-design-method.md and board-contract.md.
Glue logos, avatars, or stickers onto moving objects in a video clip so they follow the motion frame by frame — then export MP4 + GIF. Semi-supervised tracking: you read boxes off grid sheets, OpenCV CSRT carries them, and re-anchors or hand-set keyframes take over where trackers die (shot changes, walk-toward-camera scale blowups, long smooth walkaways).
When to use:
Key features:
Example usage:
"把这段视频里三只猫的头分别换成这三个 logo，做成梗图视频和 GIF"
"Cover the CEO's face with our competitor's logo in this keynote clip"
"Turn 6:30-6:50 of this bilibili video into a GIF with my avatar on the main character"
📚 Documentation: See meme-creator/SKILL.md and the bundled references/ for the tracking playbook and the asset-binding gate.
Requirements: ffmpeg; uv (bundled Python scripts carry inline dependencies). yt-dlp only when downloading from a URL.
Install:
claude plugin install data-visualization-discipline@daymade-skills
Decides whether a chart should exist and how it should be drawn before any code is written: the one conclusion each chart must support, mean vs. median, whether segments may be aggregated, stacked vs. line, dual axes, color count and cross-chart color identity, and how many columns a table keeps. Medium-agnostic — HTML reports, React/Vue dashboards, native PowerPoint charts, matplotlib/plotly/ECharts/D3.
Key features:
Five stages (intent → data validity → form → encoding → delivery): the first four end with a handoff check, the fifth is the delivery gate
A nine-item delivery gate, including a masked-render self-test and painted-span proportion checks
Reviews existing charts ("what is wrong with these charts", "why do the legends disagree")
data-visualization-discipline — full instructions. Pairs with report-with-html, which owns the report page itself.
Install:
claude plugin install report-with-html@daymade-skills
Produces the reader-facing HTML artifact — report, dashboard, architecture or journey view, data browser, or review workbench — built around the reader's question, with every number traceable and the page verified in a real browser before delivery.
Key features:
Skeleton chosen by the reader's question (how it works / what to decide / why it happened / which is better / inspect a collection)
Decision-card contract, reusable interaction components, and a warm-paper starter template
Delivery gate script with a masked (text-free) render, segment cropping for full-page review, and a zero-context independent reader review
report-with-html — full instructions. Load data-visualization-discipline for chart judgment.
Requirements: Google Chrome or Chromium; uv; Python 3.10+.
Install:
claude plugin install skill-evolver@daymade-skills
Daymade Skill Evolver assembles an improvement workflow for a specified existing Skill, using existing Skills for evidence gathering, mechanism trade-offs and authorized implementation. Reuse successful and failed work, inspect relevant competitor source, and verify changes through the target's actual consumer. Keep the current design when the evidence supports no change.
When to use: Improve a named Skill from work history, saved research or competitor mechanisms. Use Skill Creator for routine creation and editing.
Example usage:
Use skill-evolver to improve my batch-export Skill. Preserve successful exports,
investigate the interrupted-batch failure, and implement the justified repair.
Requirements: A host with access to the target source and the tools needed to edit and verify it. Resolve, load and execute available public stage Skills; local private integrations are optional. Missing required sources or data pause only dependent work.
Dependencies: This plugin ships only the orchestrator. For a missing executor, report its supported install package from the public catalog or an explicit fallback with its evidence limits.
📚 Documentation: skill-evolver.
Install:
claude plugin install app-evolver@daymade-skills
Improve an existing application from its current code and product contract, authorized successful/failed work, and competitor mechanisms. Decide what fits, implement the authorized increment and verify the actual user result.
When to use: Take the useful parts of a competitor, improve an app from past
experience, or coordinate improvements across apps with separate contracts.
Use skill-evolver when the target is a Skill.
Use app-evolver to improve my document app from this competitor. Preserve source
provenance, implement the justified improvement and verify failure recovery.
📚 Documentation: app-evolver.
Reverse Skill composition: Specialist selection and handoff.
Want to see all demos in one place with click-to-enlarge functionality? Check out our interactive demo gallery or browse the demos directory.
Use github-ops for verified PR, issue, Actions, repository, organization-access, and API operations. Use github-review-pr when a maintainer needs a current-base code review, ownership decision, or review-gated repair/landing for one contributor PR or the complete open PR queue. An author can explicitly request the same independent read-only lens for their own contribution; implementation and authorized fixes remain with github-contributor.
Combine doc-to-markdown for document conversion and mermaid-tools for diagram generation to create comprehensive documentation. Use llm-icon-finder to add brand icons.
Use deep-research to produce format-controlled research reports with evidence tables and citations. Combine with fact-checker to validate claims or with twitter-reader for social-source collection.
Use tech-selection when choosing between technologies — library, framework, storage, data format, model, build-vs-buy, or architecture. It runs a gated checklist that filters candidates by business-result anchoring, prior-art inventory, and probe-based evidence, then returns candidates + trade-offs + a recommendation when multiple survive. Combine with deep-research when the landscape itself needs research before a decision can be made.
Use competitors-analysis to discover, persist, update, and analyze competitor repositories with evidence-based source citations. Combine with deep-research when the landscape also needs broader market, pricing, or narrative research.
Use pdf-creator to convert markdown to print-ready PDFs with proper Chinese font support for formal documents and reports.
Use teams-channel-post-writer to share knowledge and statusline-generator to track costs while working.
Use the current host's native communication tools for reachable agents and sessions. Use peer-message only for uncovered local targets, cross-product/script transport, or coordination evidence that needs verification. Follow its routing contract; a denied or Held message is not a reason to switch transports.
Use repomix-unmixer to extract and validate repomix-packed skills or repositories. Use repomix-safe-mixer to package codebases securely, automatically detecting and blocking hardcoded credentials before distribution.
Use skill-creator (see Essential Skill section above) to build, validate, and package your own Claude Code skills following best practices. Use skill-evolver for a research-driven improvement round on a specified existing Skill, from relevant evidence and trade-offs to authorized changes and result verification.
Use ppt-creator to generate professional slide decks with data visualizations, structured storytelling, and complete PPTX output for pitches, reviews, and keynotes. Use slides-creator for narrative-first slide design — it guides you through the ABCDEFG storytelling framework, collects your original content first, then delegates visual generation to baoyu-slide-deck. Perfect when you have existing articles, transcripts, or talks that need to become visual slides.
Use video-comparer to analyze compression results, evaluate codec performance, and generate interactive comparison reports. Combine with youtube-downloader to compare different quality downloads.
Use youtube-downloader t
Truncated — view the full README on GitHub.
47 followers · starred May 2026
138 followers · starred May 2026
106 followers · starred Feb 2026
22 followers · starred Nov 2025