EVC

The agent workspace for marketing operators.

42 commands. 12 hooks. 3 modes. 4 starters. One setup.sh.

An opinionated marketing agent system for Claude Code. Clone it, configure it, run /project:monday to start your week.

Visual overview →

---

What this is

EVC is a folder. Inside it:

evc/
├── _templates/                 # Source of truth (tracked in git)
│   ├── commands/               # 42 command prompts with {{TOKENS}}
│   ├── hooks/                  # 12 shell scripts that fire on events
│   ├── contexts/               # 3 mode files (content/build/research)
│   ├── staging/                # Shared staging stylesheet
│   ├── SOUL.md                 # 7 operating principles
│   └── CLAUDE.md               # Top-level agent instructions
│
├── .claude/                    # Rendered output (gitignored)
│   ├── commands/               # Your 42 rendered commands
│   ├── hooks/                  # Your 12 rendered hooks
│   └── contexts/               # Your 3 rendered modes
│
├── brand/                      # Your brand brain
│   ├── canon-rules.json        # Banned phrases, USPs, required language
│   └── canon-history/          # Auto-maintained snapshots + CHANGELOG
│
├── content-engine/             # Working memory
│   ├── memory/                 # Lessons, learning log, visual log
│   ├── output/                 # Weekly posts, blog drafts, reports
│   └── calendar/               # Weekly briefs + reviews
│
├── staging/                    # Browser previews before publish
├── scripts/                    # Validation, secret scan, staging helpers
├── starters/                   # 4 industry configs (web3/saas/ecom/generic)
├── config.env.example          # Template — copy to config.env
└── setup.sh                    # One command to render everything

Every command is a markdown file. Every hook is a shell script. The workspace is just files — fork it, read it, change it.

---

Quick start

git clone https://github.com/SaigonXIII/evc.git
cd evc
./setup.sh

Setup takes 2 minutes:

1. Pick your industry (Web3, SaaS, E-commerce, Generic)
2. Fill in 5 required variables
3. Done — 42 commands installed, 12 hooks wired

Then open Claude Code in the directory:

/project:start

---

Commands

Chains (the point of entry)

Content Engine (daily social)

Blog Engine (long-form)

QA & Review

Comms

Memory & Productivity

Setup

Staging & Pipelines

---

Pipelines

Content Engine (Mon→Fri)

/intel-scan → /performance-review → /content-plan → [You review]
  → /content-week → /content-qa → [You review] → /post-today

Blog Engine (per article)

/blog-plan → [You review] → /blog-draft → /blog-qa → [You review]
  → /blog-assets → /stage → [Browser review] → /blog-publish → /repurpose

Human gates at every decision point. The system drafts, you approve.

---

The brand brain

brand/canon-rules.json is the single source of truth:

{
  "banned_phrases": [{"phrase": "...", "reason": "..."}],
  "framing_red_flags": ["best in class", "revolutionary", "..."],
  "required_language": {"product_name": "...", "company_anchor": "..."},
  "usps": ["...", "..."],
  "defunct_partners": ["OldBrand"],
  "live_partners": ["PartnerA", "PartnerB"],
  "compliance_rules": {"US": {...}, "EU": {...}, "UK": {...}}
}

Every command reads it. The canon-guardrail hook blocks violations before they ship. The canon-changelog hook snapshots every edit to brand/canon-history/. Run /project:canon-diff to see what changed and when.

---

Hooks

12 shell scripts that fire on Claude Code events. You don't call them — they run automatically.

---

Configuration

Required (5 variables)

Industry starters

Pre-filled configs in starters/:

• web3.env — crypto compliance, community-first, Discord
• saas.env — LinkedIn-heavy, case studies, product-led
• ecommerce.env — product launches, seasonal cadence, conversion
• generic.env — baseline for anything else

See config.env.example for the full variable list including CMS IDs, Linear prefix, compliance jurisdictions, and credential paths.

---

Memory

EVC gets smarter the longer you use it:

• QA failures → lessons. Failed content auto-appends to content-engine/memory/lessons.md. Ships with 14 pre-loaded.
• Performance → planning. /performance-review writes winning patterns to learning-log.md. /content-plan reads it before building next week's brief.
• Session continuity. session-end hook saves state on every stop. session-start loads it. No manual save required.
• Canon freshness. validate.sh warns if brand rules are over 30 days old.

---

Security

• .secrets/ — chmod 700, owner-only. Never committed.
• config.env — chmod 600. References token files, not inline keys.
• scripts/secret-scan.sh — Scans for leaked API key patterns. Runs on setup.
• Action audit log — every external action logged append-only to .claude/audit.log.
• Nothing auto-sends. Every post, message, and CMS push requires explicit approval.

---

Extending

Add a command

# 1. Write the prompt
vim _templates/commands/my-command.md

# 2. Render
./setup.sh

# 3. Use it
/project:my-command

Add a hook

# 1. Write the script
vim _templates/hooks/my-hook.sh && chmod +x _templates/hooks/my-hook.sh

# 2. Wire it in setup.sh → settings.json

# 3. Render + validate
./setup.sh && ./scripts/validate.sh

Every command is a markdown file. Every hook is a shell script. If you can write a prompt, you can extend EVC.

---

Credits

Built by SaigonXIII. Battle-tested internally before open-sourcing.

Inspired by Everything Claude Code — the community-built power-user kit for developers. EVC is the same idea, reshaped for marketing.

Open source under the MIT License — fork it, rename it, make it yours.