Project Harness Bootstrap

---
name: project-harness-bootstrap
description: Create or harden the baseline agent-ready structure for a repo. Use when a user asks to prepare a project for Codex, add an agent harness, standardize scripts/docs/CI, or make an existing repo easier and safer for coding agents to work in.
---

# Project Harness Bootstrap

## Purpose

Install the baseline repo interfaces that every other harness skill depends on: `AGENTS.md`, operational `docs/`, stable `scripts/`, environment examples, CI, and PR review structure.

## Inspect First

- `package.json`, `pyproject.toml`, `Cargo.toml`, `go.mod`, `Gemfile`, or equivalent manifests
- existing `AGENTS.md`, `CLAUDE.md`, `.cursorrules`, README, docs, scripts, CI, and env files
- existing test, lint, typecheck, build, and dev commands
- git status when inside a repo, so unrelated user changes are preserved

## Procedure

1. Detect the stack and current harness maturity.
   - Use manifests and existing commands, not guesses.
   - Record missing pieces and existing equivalents.
   - For large existing repos, run `harness-diagnostics` first and use its roadmap to avoid broad unfocused changes.
   - For agent-heavy or multi-session work, run or recommend `initialization-readiness-harness` and `session-continuity-harness` before implementation changes.

2. Create the baseline structure.
   - Add `AGENTS.md` as a short routing map.
   - Add `docs/00-index.md` and operational docs.
   - Add `scripts/` wrappers for stable commands.
   - Add optional surface wrappers when the repo has matching surfaces: `scripts/click-path-audit` for browser UI, `scripts/check-research-evidence` for dependency/API/security research, and `scripts/ci-local` for CI reproduction.
   - Add `.env.example` when runtime configuration exists or is expected.
   - Add CI only when a repo uses GitHub or the user requested GitHub Actions.

3. Preserve useful generated and user-authored files.
   - If a file already exists, merge the harness intent into it.
   - Do not overwrite project-specific instructions, scripts, or docs without reading them.

4. Delegate details.
   - Use `agents-md-author` for `AGENTS.md`.
   - Use `repo-knowledge-base-builder` for docs.
   - Use `quality-gates-installer` for scripts and CI checks.
   - Use `local-runtime-harness` for env/runtime setup.
   - Use v2 lifecycle skills for deeper security, dependency, docs sync, observability, migration, release, and review workflows.
   - Use `feature-state-machine` and `termination-gatekeeper` when the repo needs WIP control or completion evidence.
   - Use `project-local-skill-generator` after the harness exists to add thin repo-specific `.codex/skills` without copying plugin skills.

## Required Stable Interface

Prefer these repo commands when feasible:

- `./scripts/dev`
- `./scripts/lint`
- `./scripts/typecheck`
- `./scripts/test`
- `./scripts/build`
- `./scripts/validate`
- `./scripts/check-boundaries`
- `./scripts/check-docs`
- `./scripts/audit-entropy`

Optional wrappers should be generated only when backed by repo artifacts or current changed surfaces:

- `./scripts/test-e2e`
- `./scripts/click-path-audit`
- `./scripts/check-research-evidence`
- `./scripts/ci-local`

Scripts may be thin wrappers around project-native commands. If a command does not apply, the wrapper should explain why and exit successfully only when that absence is intentional.

## Files To Create Or Update

- `AGENTS.md`
- `docs/00-index.md`
- `docs/development/commands.md`
- `docs/development/environment.md`
- `docs/architecture/overview.md`
- `docs/testing/strategy.md`
- `docs/quality/code-style.md`
- `scripts/*`
- `.github/workflows/ci.yml`
- `.github/pull_request_template.md`
- `.codex/skills/*/SKILL.md` only when generated by `project-local-skill-generator`

## Validation

- Run the narrowest new wrappers first.
- Run `./scripts/validate` after the baseline is complete.
- Check that docs link to real paths and commands.
- Check that generated scripts are executable.

## Completion Criteria

- A new agent can answer what the project is, how to run it, how to test it, where code belongs, what patterns to copy, and how to validate changes.
- No generated file contains placeholder TODO text.
- Existing project-specific files are preserved or intentionally merged.
- Multi-step agent work has initialization, continuity, feature-state, or termination guidance when useful.