Gem Team

Turn AI coding into an engineering process.

Agent definitions that enforce good software engineering: optimizing cost, time, and quality.

Visit Homepage

The Problem

Current AI coding is often one-off and ad-hoc. You get code, but you don't get a repeatable process. This leads to inconsistent quality, wasted tokens, and a lack of long-term learning.

The Solution

Gem Team wraps your AI with a disciplined engineering delivery system. It enforces good software engineering practices automatically, so you get better results with less effort.

Why Gem Team?

• Quality by Default: TDD and acceptance checks always apply; reviews and security audits run when risk requires them. No more "vibe coding" that breaks in production.
• Smart & Efficient: 40-60% less context per task through scoped handoffs and proportional architecture. Cached tokens and compact evidence paths keep costs predictable and your AI focused.
• Works With Your Tools: Seamless integration with Copilot, Claude, Cursor, Codex, Gemini, and Windsurf. Use your preferred environment.
• Learns & Improves: Remembers what works and extracts reusable skills. Your AI gets smarter and more efficient over time.
• Resumable Plans: Every MEDIUM/HIGH task gets a persistent plan ID. Pause, resume, or extend work without losing context or re-discovering what you already know.
• Works With Any Model: Hardened output contracts and relational invariant fallbacks mean every agent works across commercial and open models — not just the ones that memorized your schema.

Intelligent Model Routing

Gem Team automatically uses the right model for each kind of work:

• Premium models handle planning, debugging, and review where deeper reasoning matters.
• Explore models handle research, implementation, testing, documentation, and other bounded tasks efficiently.
• Configurable tiers let you choose the models and providers that fit your budget and workflow.

This gives you stronger verification where it matters without paying the highest model cost for every task. Configure it once in .gem-team.yaml:

model_routing:
  enabled: true
  tiers:
    premium: "your-strong-model (provider)"
    explore: "your-fast-model (provider)"

TL;DR: Gem Team turns AI coding into a structured, repeatable engineering process with built-in quality, efficiency, and learning.

Quick Start

Install APM first:

# macOS / Linux
curl -sSL https://aka.ms/apm-unix | sh

# Windows PowerShell
irm https://aka.ms/apm-windows | iex

# Verify
apm --version

Install Gem Team into your current project:

apm install mubaidr/gem-team --target copilot,claude,cursor,opencode,codex,gemini,windsurf

Or install for one target only:

apm install mubaidr/gem-team --target copilot

Install globally for personal use:

apm install -g mubaidr/gem-team

APM records the resolved commit in apm.lock.yaml. Repeating apm install replays that lockfile; it does not silently upgrade an existing installation. Refresh Gem Team explicitly when desired:

# Project-scoped installation
apm update mubaidr/gem-team --yes

# Global installation
apm update -g mubaidr/gem-team --yes

To check for an update to the APM CLI itself, use apm self-update --check.

For reproducible environments, pin a release tag:

apm install 'mubaidr/gem-team#gem-team-v<version>' --target copilot

Replace <version> with a published version from the GitHub Releases page.

After the first install, commit the generated APM files that belong to your repo, especially apm.yml, apm.lock.yaml, and the generated harness directories such as .github/, .claude/, .cursor/, .opencode/, .codex/, .gemini/, or .windsurf/. Do not commit apm_modules/.

APM can auto-detect targets from existing harness directories, but explicit --target is recommended for predictable installs and fresh repositories.

Direct Git installs use the canonical sources in .apm/. Maintainers do not need to commit build/; release archives and checksums are generated and attached automatically to each GitHub Release.

The Process

Gem Team uses a structured workflow to turn AI coding into a reliable engineering process:

1. Route: Classify the request from supplied evidence and select only the workflow depth it needs.
2. Plan: Use an in-memory wave plan for TRIVIAL/LOW work or a persistent, planner-confirmed wave plan for MEDIUM/HIGH work.
3. Build: Execute every wave plan through the same ordered-wave loop, using TDD and specialist agents.
4. Verify: Check every task and run reviewer integration checks only when changed-scope risk requires them.
5. Learn: Promote only stable, high-confidence patterns after successful execution.

Features

• Risk-Based Quality Gates: TDD and deterministic verification always apply; specialist reviews and audits run when the plan or changed scope requires them.
• Effortless Context: Progressive context management prevents bloat. Scoped handoffs, bounded planning_context, and evidence-by-reference keep each agent's token footprint minimal while maximizing cached token reuse across waves.
• Smart Routing: Tasks are automatically routed to the right agents based on complexity.
• Parallel Execution: Independent tasks run in parallel within waves; overlapping ownership is serialized to prevent conflicts.
• Resumable Plans: Every MEDIUM/HIGH task gets a persistent plan ID and plan.yaml. Pause, resume, or extend work without losing context.
• Reusable Knowledge: High-confidence patterns and skills are extracted and reused for future tasks.
• Cost Efficiency: Model routing, output hygiene, and compact handoffs ensure you only use the tokens you need. Evidence travels by reference, not by copy — keeping context usage low and cache hits high across waves.
• Failure Classification: Every failure is classified (retry, fixable, replan, flaky, regression, platform-specific, test-bug) so the Orchestrator routes it to the right agent instead of blindly retrying.
• Verification Boundary: The Orchestrator never re-verifies or second-guesses specialist output. Verification is owned exclusively by the specialist responsible for the work.

How it Works

Gem Team installs a set of specialized agents that work together under the guidance of an Orchestrator. This team follows a disciplined workflow that includes planning, implementation, verification, and learning.

• Specialist Agents: Dedicated agents for planning, research, implementation, review, and more.
• Orchestration: One wave loop coordinates ordered work, parallel tasks, bounded retries, and final acceptance checks at every complexity level.
• Context Management: Execution agents receive an authoritative task_definition with a nested handoff; constraints, evidence, and prior-wave outputs travel through it. Planner receives a bounded planning_context; reviewer uses a dedicated review handoff; every delegate receives only a role-scoped configuration snapshot.

Agent Roles

Compatible Tools

Gem Team works with your favorite AI coding tools:

Configuration

Gem Team is designed to work out of the box with smart defaults. You can customize behavior by editing the AGENTS.md file or specific agent definitions in the .apm/agents/ directory.

Reviewer and critic modes

gem-reviewer uses three independent axes:

• review_mode: standard, high, or critic controls review intensity.
• review_target: plan, task, code, decision, docs, config, or integration selects what is reviewed.
• review_scope: changed, affected, or full limits the evidence breadth.

TRIVIAL/LOW work does not invoke the planner or reviewer during planning. MEDIUM/HIGH work receives one pre-execution plan review: standard for MEDIUM, high for HIGH or high-risk work, and critic for architecture, breaking-change, or cross-domain signals. Later integration review is risk-triggered, not a routine wave gate.

Discussion is answered directly. A requested evaluation or decision becomes a read-only challenge with review_mode: critic, review_target: decision, and review_scope: full. Critic mode does not mutate files or claim implementation. Its subject and context are passed through handoff:

review_mode: critic
review_target: decision
review_scope: full
handoff:
  critic_subject:
    objective: str
    proposal: str
    constraints:
      - str
    alternatives:
      - str
    evidence:
      - str
    decision_needed: str
  critic_context:
    audience: str
    time_horizon: str
    success_criteria:
      - str
    known_unknowns:
      - str

Learn More

• Documentation
• Contributing
• License

Support

If you have questions or need help, please open an issue on GitHub.