Skills
Pre-built workflows agents can invoke as slash commands.
Available skills
| Skill | What it does |
|---|---|
/build | Read commands.yaml → run install, build, test, lint |
/review | git diff → tests → decisions.md check → maturity |
/bugfix | Reproduce → find root cause → fix → verify → regression test |
/delegate | Save session → create swarm task → handoff to another agent |
/feedback | Log bug / suggestion / friction / positive observation |
/status | Doctor + validate + maturity + swarm status |
How they work
Each skill reads commands from ai/manifest/commands.yaml — never hardcodes project-specific commands. Skills also reference MCP servers when appropriate:
- context7 — for unfamiliar library APIs or version errors
- sequential-thinking — for complex multi-step debugging
Delegation rules in skills
Each skill includes escalation guidance:
- Stuck after 2 attempts → escalate to opus model
- Need to search code → delegate to haiku
- Architecture change needed → delegate to architect agent