Portable AI agent workflow skills. Define how your AI coding agent commits, reviews, audits security, and closes epics — consistently across any project.
Works with Claude Code, Codex, Cursor, and OpenClaw.
Every developer using AI coding agents repeats the same workflows: formatting commits, reviewing PRs, auditing security, closing epics. Without a standard, each session starts from scratch.
This kit gives you battle-tested, reusable skills that define exactly how your agent should behave — install once, use everywhere.
npx @kanam/workflow-kitThis will:
- Ask which skills you want (defaults recommended)
- Ask for your project's scopes (for commit messages)
- Install everything to
.agents/skills/
| Skill | Command | What it does |
|---|---|---|
| commit | /commit |
Format commits with Conventional Commits, project scopes, and Co-Authored-By |
| commit-formatter | (auto) | Analyzes diff and writes formatted commit message |
| pr | /pr |
Create structured PRs with description, review, and merge |
| pr-review | /review |
Cognitive pre-merge linter — checks correctness, tests, conventions, security, performance |
| new-task | /new-task |
Initialize tasks: read spec → create branch → implement → merge |
| refactor | @refactor |
Refactor without changing behavior |
| doc-writer | @doc-writer |
Generate and update technical documentation |
| Skill | Command | What it does |
|---|---|---|
| copilot | (auto) | Universal orchestrator — Pre-Flight checklist + 5 Gates per task |
| close-epic | /close-epic |
Formal epic closure: security audit, CHANGELOG, tag, merge |
| security-audit | /security-audit |
Threat model: multi-tenancy, auth guards, secrets, data leaks |
| product-reset | /product-reset |
Reset estratégico: diagnosticar, visionar, planificar, ejecutar |
| docs-sync | /docs-sync |
Detect and update out-of-sync docs when closing epics |
Each skill is a SKILL.md file in .agents/skills/<name>/. AI coding agents that support custom skills (Claude Code, Codex, Cursor, OpenClaw) read these files and follow the defined workflow.
When the copilot skill is active, every task follows this cycle:
Pre-Flight → Gate 0: Branch → Gate 1: Impl → Gate 2: Build → Gate 3: CHANGELOG → Gate 4: Commit + Review
Pre-Flight: Before writing a line of code, the agent verifies context is complete (CLAUDE.md read, correct branch, dependencies confirmed). Won't advance without explicit confirmation.
The 5 Gates: Each gate is a real block. If the build fails at Gate 2, the agent stops and fixes before continuing. If the review at Gate 4 returns blocking changes, no merge happens.
The commit skill has a {{FILL}} placeholder for project scopes. When you run npx @kanam/workflow-kit, it asks for them. You can also edit .agents/skills/commit/SKILL.md directly.
If your project uses AI agent commits, the commit and commit-formatter skills have a {{FILL}} for the Co-Authored-By footer. Edit the SKILL.md to add your agent's name and email.
The pr skill has {{FILL: owner}} and {{FILL: repo}} placeholders for the GitHub API calls. Edit .agents/skills/pr/SKILL.md with your values.
The docs-sync skill has a mapping table with {{FILL}} placeholders for your project's file-to-doc patterns. The first run will auto-discover them.
| Agent | How it reads skills | Status |
|---|---|---|
| Claude Code | .agents/skills/<name>/SKILL.md |
✅ Tested |
| Codex | .agents/skills/<name>/SKILL.md |
✅ Compatible |
| Cursor | .cursor/skills/<name>/SKILL.md |
|
| OpenClaw | Workspace skills via skill_workshop |
✅ Compatible (see integration guide) |
.agents/
├── skills/
│ ├── copilot/SKILL.md
│ ├── commit/SKILL.md
│ ├── commit-formatter/SKILL.md
│ ├── pr/SKILL.md
│ ├── pr-review/SKILL.md
│ ├── new-task/SKILL.md
│ ├── close-epic/SKILL.md
│ ├── security-audit/SKILL.md
│ ├── product-reset/SKILL.md
│ ├── refactor/SKILL.md
│ ├── doc-writer/SKILL.md
│ └── generic-docs-sync/SKILL.md
└── antigravity-bridge.md
MIT — do whatever you want with the code.
PRs welcome. For major changes, open an issue first.