# Agent Project Bootstrap

> Bootstrap the Agent Project Playbook into a user-selected project with a dry-run first. Use when the user explicitly asks to initialize project collaboration rules, BLUEPRINT, ARCHITECTURE, CURRENT_HANDOFF, or the project-local lifecycle, multi-agent, and docs skills.

- Skill: `candycong2001-ui/agent-project-bootstrap` (Agent Skill)
- Install (CLI): `npx skillmds@latest add candycong2001-ui/agent-project-bootstrap`
- Raw SKILL.md: https://api.skillmd.com/api/skills/candycong2001-ui/agent-project-bootstrap/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: candycong2001-ui (https://skillmd.com/u/candycong2001-ui)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/candycong2001-ui/agent-project-bootstrap

---


# Agent Project Bootstrap

## Procedure

1. Confirm the target project root, project name, and document locale (`en` or `zh-CN`).
2. Use the `simple` profile unless current repository evidence shows a live runtime, shared state, real users, credentials, or external writes; then recommend `production`.
3. Run `bin/meminit <path> <name>` from this repository. Add `--production` only for the production profile, or `--with-projectmem` only when the optional cold-memory adapter is explicitly wanted.
4. Review the dry-run report; do not make the user remember the command.
5. If the target already has Playbook files, run the same command with `--upgrade`; apply the plan only when it has no blocked managed files.
6. If the original request explicitly authorizes installation and the dry-run has no conflict, apply without another confirmation. Otherwise ask once before `--apply`.
7. Activate the generated documents using current repository evidence and keep `CURRENT_HANDOFF.md` limited to one current state.
8. Mark unverified commands, runtime side effects, and risks instead of guessing.
9. Run the installed Docs Lint before handing substantial implementation work back.

## Safety

- Never overwrite existing project files.
- Never import another project's memory or status.
- Never store secrets / 密钥, tokens, cookies, personal data, or private logs.
- Never configure MCP, cross-project memory, background watchers, or global Agent settings from this skill.
- `--upgrade` is read-only unless explicitly combined with `--apply`; safe upgrade updates metadata, restores missing managed files, and stops when a managed file was customized.
- Never overwrite `AGENTS.md`, BLUEPRINT, ARCHITECTURE, or CURRENT_HANDOFF during upgrade.
- The default simple install does not include production safety tooling.
- The default install is Playbook Core only and does not require `pjm`.
- The optional projectmem adapter keeps hooks, watcher, history backfill, global inheritance, and automatic capture off.
- Current repository evidence outranks every generated placeholder.