Bootstrap Project

Bootstrap A Project For Codex

Run the Codex-oriented new-repo setup workflow in one guided pass. This is a thin orchestrator: it owns sequencing and handoffs, and delegates detailed work to the focused skills.

Sequence:

AGENTS.md setup -> where-agents-md -> setup-permissions -> propose-automation -> setup-ci optional -> obsidian-vault optional -> durable memory suggestions.

Operating Principles

1. Orchestrate, do not reimplement delegated skills.
2. Pause after each step and continue only after user approval.
3. Be idempotent: detect existing AGENTS.md, .codex/, .agents/, and repo config.
4. Prefer shared repo guidance only when it is useful to teammates; keep personal Codex

config personal.

Procedure

0. Orient

• Confirm the working directory is the intended repo root.
• Detect existing setup: root/nested AGENTS.md, AGENTS.override.md, .codex/,

.agents/, CLAUDE.md, README, CI, and project manifests.

• For a large repo, delegate broad exploration to a read-only subagent when available.
• Show the planned sequence and let the user drop steps.

1. Root Codex Instructions

• If no root AGENTS.md exists, propose creating one from observed repo facts.
• If one exists, offer to review/update it instead of regenerating.
• If CLAUDE.md exists, use it as source context but do not copy Claude-only commands,

hook syntax, or permission syntax blindly.

2. Scoped Instruction Placement

• Invoke where-agents-md to identify high-value nested AGENTS.md candidates.
• Present recommendations and create files only for directories the user approves.

3. Codex Permissions And Config

• Invoke setup-permissions to review Codex sandbox, trusted project, and command

approval expectations.

• Do not edit global ~/.codex/config.toml unless the user explicitly asks. Prefer

project-scoped guidance or a patch proposal.

4. Automation Proposal

• Invoke propose-automation to suggest project Codex skills, custom agents, hooks, or

plugin packaging.

• Scaffold only approved items.

5. CI Pipeline

• Optionally invoke setup-ci for GitHub Actions. Skip if the repo is not on GitHub or

if the user declines.

6. Obsidian Vault

• Optionally invoke obsidian-vault if the user wants a repo knowledge base.

7. Durable Memories

• Offer only genuinely durable, non-secret facts as project memories. Do not record

information already captured in AGENTS.md or README.

8. Wrap Up

• Summarize created/changed files.
• Remind the user that new Codex skills and agents are picked up in a new Codex session.

Guardrails

• Never run the full chain unattended.
• Never clobber existing setup; merge or propose targeted edits.
• Keep Claude and Codex artifacts separate unless the user explicitly asks for a shared

compatibility file.

• Never fabricate repo facts.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: ada-ggf25
• Source: ada-ggf25/AI-Tools
• License: MIT

Install and usage instructions live in the source repository linked above.