An open API service indexing awesome lists of open source software.

https://github.com/dd3ok/onboardkit

A lightweight Agent Skill for initializing and maintaining AGENTS.md and repository docs.
https://github.com/dd3ok/onboardkit

agent-skills agents-md codex developer-tools documentation

Last synced: 1 day ago
JSON representation

A lightweight Agent Skill for initializing and maintaining AGENTS.md and repository docs.

Awesome Lists containing this project

README

          

# onboardkit

onboardkit is a lightweight Agent Skill designed primarily for Codex to maintain `AGENTS.md` and agent-facing repo docs. Other Agent Skills runtimes are format-compatibility targets, not verified support.

It helps an agent decide:

- how to initialize a small `AGENTS.md`
- how to respect active overrides, fallback instructions, and nested scopes
- how to fill discovered facts while avoiding invented repo-specific guidance
- when to ask the user or report Needs Input instead of guessing
- what belongs in always-on `AGENTS.md`
- when to create `docs/README.md` and deeper docs
- what stale, duplicate, generated, or low-signal docs should be removed
- how README and agent-facing docs should stay aligned

There is no helper command, npm package, scaffolder, runner, dashboard, or task ledger in this repository.

## Install

For Codex, ask the built-in installer to copy only the canonical runtime directory:

```text
Use $skill-installer to install onboardkit from https://github.com/dd3ok/onboardkit/tree/main/skills/onboardkit
```

Restart Codex if the installed skill does not appear. This route excludes repository-only docs, CI, and eval fixtures from the installed skill.

For other runtimes, copy only `skills/onboardkit/` into that runtime's current official user-level skills directory. Those runtimes remain format-compatibility targets, not verified support.

Claude Code can use the Agent Skills format but does not automatically discover `AGENTS.md`; target repositories need `CLAUDE.md` routing, such as `@AGENTS.md`, for those instructions to apply.

To update an installer-managed copy, remove its installed `onboardkit` directory and repeat the installer prompt.

Existing full-repository installations remain compatible through the root `SKILL.md` entrypoint. Update those legacy checkouts with:

```bash
git -C ~/.agents/skills/onboardkit pull --ff-only
```

Do not use a full-repository checkout for new installations. Clone normally outside a skills directory for source development or repository validation.

Then ask your coding agent in a target repository:

```text
Use onboardkit to clean up this repo's AGENTS.md and docs.
```

Implicit invocation is enabled. Invoke `$onboardkit` explicitly for cleanup. Deletion is authorized only when the request says delete/remove and supplies each literal file path or a file-matching glob; the agent must not broaden it, and a directory path alone is insufficient.

Common prompts:

```text
Use $onboardkit to initialize lightweight AGENTS.md and docs routing for this repo.
Use $onboardkit to refresh this existing AGENTS.md, preserving it and changing only what evidence or routing requires.
Use $onboardkit to audit AGENTS.md and agent-facing docs for stale or duplicate guidance without deleting files.
Use $onboardkit to do a monthly maintenance pass on agent-facing docs.
```

## Files

```text
.github/workflows/validate.yml repository-only static validation
skills/onboardkit/ canonical runtime skill
SKILL.md legacy compatibility entrypoint
agents/openai.yaml legacy UI metadata
evals/evals.json behavior evaluation cases
evals/eval_queries.json trigger boundary cases
evals/files/ isolated behavior fixtures
AGENTS.md maintenance guidance for this repo
README.md user-facing summary
LICENSE source license
.gitignore local-file exclusions
.gitattributes Git text normalization
```

## Maintenance

Keep the skill instruction-only unless a future requirement clearly needs deterministic tooling.

Use the [AGENTS.md](AGENTS.md) Definition of Done as the canonical maintainer checklist. Keep runtime instructions and metadata under `skills/onboardkit/`; root skill files exist only for legacy full-checkout compatibility. Eval JSON and fixtures remain declarative, not a model runner; fresh release validation copies fixtures into isolated workspaces and compares current with baseline.

## References

- [OpenAI Codex skills](https://learn.chatgpt.com/docs/build-skills)
- [OpenAI AGENTS.md guidance](https://learn.chatgpt.com/docs/agent-configuration/agents-md)
- [OpenAI Codex best practices](https://learn.chatgpt.com/guides/best-practices)
- [Claude Code memory guidance](https://code.claude.com/docs/en/memory)
- [Claude Code skills](https://code.claude.com/docs/en/skills)
- [GitHub Copilot repository instructions](https://docs.github.com/en/copilot/how-tos/copilot-on-github/customize-copilot/add-custom-instructions/add-repository-instructions)
- [Agent Skills specification](https://agentskills.io/specification)
- [Agent Skills evaluation guide](https://agentskills.io/skill-creation/evaluating-skills)
- [AGENTS.md](https://agents.md/)

## License

MIT