agent-skills

Skills for Claude Code and Codex.
The durable loops drive a repository toward a goal. The parallel skills split a review, an
investigation, a research question or a change across independent agents. The rest cover
shipping and growing a repository, replying, recovering cut-off work, and explaining.

Features

• /investigate, /review-pr, /review-code, /research and /implement split the work across parallel agents, scaled to the size of the task, and check every finding before acting on it.
• /sergio-loop runs a durable goal loop in one repository, one small slice at a time, with goals and evidence in docs/ and resumable state in .omc/sergio-loop/.
• An optional global Stop hook continues the loop in the same session and stays inert everywhere else.
• /refine-loop makes small, behavior-preserving improvements ranked by ROI = Impact × Confidence ÷ Effort, one per round, each checked by a fresh reviewer.
• /docs-loop audits documentation against the default branch, as a dry run, with staged edits, or with one pull request per repository. It never merges.
• None of them treats a stored goal as permission to merge, publish or deploy.
• /kb-research and /kb-review answer and review from your team's own record, through one adapter file you fill in for whatever knowledge base you have.
• /grow-my-repo plans where an open-source repository gets listed (GitHub itself, registries, directories, public catalogues, media, communities, funding) and does the submissions an agent may do. /submit-awesome sends one entry per awesome list, by each list's rules. Both hand the maintainer a packet wherever a place wants a human.
• /file-pr opens pull requests whose title says why the change matters and whose body starts with the problem. /verified-reply drafts a reply you send as yourself, every claim checked against a live source first.
• /revive brings work cut off by a usage limit, a crash or a compaction back from where it really stopped, and checks the world before every write so nothing happens twice.
• /eli5 explains work or a system in plain words with real names, and /help-decide lays out a decision so one word answers it.
• A validator checks every skill package, and checks each loop against the shared loop contract.

Quick start

git clone https://github.com/GeiserX/agent-skills.git && cd agent-skills
mkdir -p ~/.claude/skills && ln -s "$PWD/skills/sergio-loop" ~/.claude/skills/sergio-loop

Then run /sergio-loop <goal> in a fresh Claude Code session. If a skill of the same name is already installed, the installer moves it aside first and also installs the Stop hook. In Codex, link from codex/skills/ into ~/.agents/skills/ and call a skill as $<name>; the Stop hook is Claude Code only.

Documentation

The full documentation is at geiserx.github.io/agent-skills.

• Getting started: the skill and hook installers, settings.json entries, canonical sources and local overlays
• Usage: what each skill does, its defaults and invocations, the goal loop, the parallel skills, the knowledge-base skills, and the skills for growing a repository, pull requests, replies and recovery
• Development: validating the skill packages, rebuilding the Codex copies, running the tests
• Related projects: the skills from other projects kept in vendor/, and the repository this one replaced

Other people's skills

vendor/ holds skills written by other people, kept here as pinned snapshots with their licences. They are not installed by default; link the ones you want the same way as the others.

• pstack by Lauren Tan (MIT), 52 skills such as unslop, blast-radius and show-me-your-work, taken from Michael Denyer's Claude Code port. Sources, commits and local changes are in vendor/pstack/PROVENANCE.md.
• writing-for-agents from Matt Pocock's skills (MIT). See vendor/mattpocock-skills/PROVENANCE.md.
• frontend-design from Anthropic's skills (Apache-2.0), for page and UI design. See vendor/anthropic-skills/PROVENANCE.md.

License

GPL-3.0-or-later