| name | harness-setup |
| description | First-time project scaffold for agentm. Populates init.sh with real commands, verifies it boots, seeds AGENTS.md, features.json, and the wiki/ scaffold. Run once per project. Prefixed harness- to avoid collision with Codex built-ins. Does not plan — planning is the harness-plan skill. Use when this capability is needed. |
| metadata | {"author":"alexherrero"} |
harness-setup skill
Run the setup phase of agentm. Full spec: harness/phases/01-setup.md. Read it and follow it.
Non-negotiable constraints
- Inventory before interviewing. Read
README.md, package.json/go.mod/etc., .github/workflows/, any existing AGENTS.md. Ask only what the inventory can't answer.
- Populate
.harness/init.sh with real commands, not guesses. If unsure, ask. A broken init.sh breaks every later phase.
- Verify
init.sh boots cleanly — run it and confirm exit 0 before finishing.
- Do not invent features for
.harness/features.json. Empty is fine.
- Merge, don't overwrite existing
AGENTS.md. It may contain project-specific content.
- No planning.
harness-setup is pure scaffolding. Planning is the harness-plan skill.
- Dispatch the
documenter subagent after boot verification to populate the wiki/ seed pages (Getting-Started, Runbook, Product-Intent, Overview, Home, _Sidebar) from the inventory and interview. Resolve any OPEN QUESTIONS before finishing.
- Offer GitHub Project creation (user-scoped,
gh project create --owner @me) — ask first. If accepted, write .harness/project.json.
Start by reading what's in the project now, then interview briefly on anything the inventory didn't settle.
Source: alexherrero/dev-setup — distributed by TomeVault.
