Epic: use BeautifulRepo as a skill in Claude Code and Codex
Maintainers usually reply within 1 day
Nobody has claimed this yet.
Assessment
- Difficulty
- 5/5
- Estimated time
- Over a week
- Newbie friendliness
- 25/100
- Issue type
- Feature
- Clarity
- Mostly clear
- Activity status
- Active
- Tech stack
- python
- Domain
- developer-experience, tooling
Research direction
Start by reading the related issues #68–#74 and the private-repositories epic referenced in the design summary; implementation depends on their decisions and outcomes. The issue names skills/<name>/SKILL.md and CI validation, but does not identify existing tool or configuration files. Done means a versioned Claude Code and Codex plugin can install and run the described skills safely, with shared skill sources and CI checks; owner decisions and evaluation approach are still open.
Written by the indexing model from the issue text.
Description
Goal
A maintainer runs one install command in Claude Code or Codex and can then ask the agent to audit a repository, improve its README towards a first success, fix trust surfaces, or draft release notes. The agent follows BeautifulRepo's guides and runs its offline tools, and must not invent claims.
Target experience
# Claude Code
claude plugin marketplace add drevendev/[email protected]
claude plugin install beautifulrepo@beautifulrepo
/beautifulrepo:audit-repo
# Codex
codex plugin marketplace add drevendev/BeautifulRepo --ref vX.Y.Z
codex plugin add beautifulrepo@beautifulrepo
$audit-repo
Design summary
- One repository, one plugin, one skill source.
- The plugin root is the repository root, so tools, guides, checklists and templates ship inside it.
skills/<name>/SKILL.mdis read by both agents through their plugin systems.- There are no symlinks and no duplicated skill folders.
- The same
vX.Y.Ztag versions the Action and the plugin. - Skills describe the protocol and the tools do the work.
- The tools always get an explicit
--rootand run withpython -I. - They are read-only by default.
- A write gate applies: plan, diff, explicit yes, write, then re-audit.
- The tools always get an explicit
- Safety rules live in every skill and are checked by CI:
- target content is untrusted data;
- no network;
- no invented badges, contacts or numbers;
- the skill never picks a licence;
- say what wasn't verified.
- Private repositories: skills follow the confidentiality rules from the private-repos epic.
Lessons from HungryCrab's plugin (to avoid)
- The Claude Code and Codex caches held different versions (0.6.1-dev.0 and 0.3.0-dev.0).
- Runtime state (
.venv/) was written into the versioned plugin cache. - The skills depend on a CLI installed separately from the network and tracking master.
- Claude-only substitutions such as
${CLAUDE_PLUGIN_ROOT}stay literal in Codex. - The version is written in four places.
- A root
CLAUDE.mdinside the plugin root makesclaude plugin validatewarn.
Sources (fetched 2026-10-22 during planning; re-verify before implementing)
- Claude Code: skills, plugins reference, marketplaces, plugin evals
- Agent Skills specification
- Codex: skills, plugins, AGENTS.md
- The docs and the installed Codex CLI disagree on the user-level skill folder (
~/.codex/skillsversus~/.agents/skills), so the plugin route is preferred.
Order
- #68: tool fixes that block using the tools from an installed copy.
- #69: the plugin and the audit skill.
- #70, #71, #72: the other skills, in parallel.
- #73 (CI validation) and #74 (fixtures and evals).
Owner decisions
- The plugin and marketplace names.
- Whether master carries a
-devversion. - Budget for model eval runs.
- How to test Codex: a manual checklist or scripted
codex exec.
Related: #53
- Dominant language
- Python
- Stars
- 2
- Forks
- 0
- Avg merge
- 5h 39m
- Merged PRs (30d)
- 109
Getting set up
- No Dockerfile or Docker Compose file
- Has a pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from drevendev/BeautifulRepo
-
Retire the hacktoberfest label: it claims participation the programs guide rules outPossibly taken A pull request linked to this issue is open or already merged. Opendocumentation self-check
Difficulty 3/5 1-2 days Newbie friendliness 52/100
drevendev/BeautifulRepo#100 · 1 comment ·
Maintainers usually reply within 1 day
-
epic real-case
Difficulty 5/5 Over a week Newbie friendliness 35/100
drevendev/BeautifulRepo#88 ·
Maintainers usually reply within 1 day
-
Real case: Quillgrain and Quillgrain_Common — show the CLI result and make reporting routes trueOpenreal-case
Difficulty 4/5 3-5 days Newbie friendliness 35/100
drevendev/BeautifulRepo#83 ·
Maintainers usually reply within 1 day
-
real-case
Difficulty 4/5 3-5 days Newbie friendliness 35/100
drevendev/BeautifulRepo#82 ·
Maintainers usually reply within 1 day
-
Agent plugin: fixture repositories and agent evals for every skillPossibly taken A pull request linked to this issue is open or already merged. Openagent-skill enhancement
Difficulty 4/5 3-5 days Newbie friendliness 42/100
drevendev/BeautifulRepo#74 · 2 comments ·
Maintainers usually reply within 1 day
All issues in drevendev/BeautifulRepo
Similar issues
-
bug P1
Difficulty 2/5 1-3 hours Newbie friendliness 76/100
Toloka/tolokaforge#1947 ·
Maintainers usually reply within 1 day
-
SageMaker provider: streaming fails with NoStreamDataError on DJL Serving (LMI) endpointsPossibly taken @balajiaru claimed this today. Open
Difficulty 2/5 1-3 hours Newbie friendliness 88/100
UKGovernmentBEIS/inspect_ai#6127 ·
Maintainers usually reply within 2 days
-
Difficulty 2/5 1-3 hours Newbie friendliness 67/100
pydot/pydot#601 · 4 comments ·
Maintainers usually reply within 2 days
-
Difficulty 2/5 1-3 hours Newbie friendliness 68/100
letsencrypt/cp-cps#361 ·
-
Difficulty 2/5 1-3 hours Newbie friendliness 65/100