agent-skills
AGENTS.md4 starsChanged 20 days agoMITView on GitHub
This repository publishes agent skills as Firzus/agent-skills. Each skill lives in skills/<section>/<name>/, or skills/game/unity/<name>/ for the Unity skills, where the section is stack, engineering, web, or game. stack holds the development stack entered through /work; the other sections hold standalone skills. Two skills rely on an application kept in its own workspace: mcp/tauri-agent-kit/ (Windows diagnostics for the tauri skill) and mcp/reverse-engineer/ (MCP server for the reverse-engineer skill).
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository39 other files it gives its agents.
- authoringSkill
- chrome-devtoolsSkill
- gamificationSkill
- imagegenSkill
- improve-architectureSkill
- instructionsSkill
- monorepoSkill
- reverse-engineerSkill
- runbookSkill
- security-reviewSkill
- setup-harnessSkill
- skillsSkill
- unslopSkill
- video-reportSkill
- youtube-transcriptSkill
- code-reviewSkill
- debugSkill
- interviewSkill
- prototypeSkill
- researchSkill
- workSkill
- adsenseSkill
- astryxSkill
- cloudflare-cliSkill
- dokploy-best-practicesSkill
- extract-themeSkill
- nextjsSkill
- payload-cmsSkill
- shadersSkill
- swrSkill
- tanstack-querySkill
- tanstack-routerSkill
- tanstack-startSkill
- tanstack-storeSkill
- tauriSkill
- vite-plusSkill
- web-assets-optimizationSkill
- web-extensionSkill
- zodSkill
The file
We split this AGENTS.md into the rules it gives an agent
4 rules. About 0 are common, 0 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 9
Write and revise skills with skills/engineering/authoring/SKILL.md.
- Very specific: line 10
Write and revise project instructions with skills/engineering/instructions/SKILL.md, including its approval gate for changes to this file.
- Very specific: line 14
When adding, renaming, moving, or removing a skill, update together its folder, its section's group in .claude-plugin/marketplace.json, and …
- Very specific: line 19
Keep each application's manifests, lockfile, dependencies, build, and tests inside its mcp/<name>/ workspace, with generated output, …
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AGENTS.md
Project
This repository publishes agent skills as Firzus/agent-skills. Each skill lives in skills/<section>/<name>/, or skills/game/unity/<name>/ for the Unity skills, where the section is stack, engineering, web, or game. stack holds the development stack entered through /work; the other sections hold standalone skills. Two skills rely on an application kept in its own workspace: mcp/tauri-agent-kit/ (Windows diagnostics for the tauri skill) and mcp/reverse-engineer/ (MCP server for the reverse-engineer skill).
Skills
- Write and revise skills with
skills/engineering/authoring/SKILL.md. - Write and revise project instructions with
skills/engineering/instructions/SKILL.md, including its approval gate for changes to this file. - Stack and engineering skills assume that
skills/engineering/setup-harness/references/operating-policy.mdis installed. A general rule lives only in that policy; a skill states its task-specific actions and keeps its own safety boundaries (commit, merge, deployment, force-push, external writes, secrets). - Stack skills call each other by relative path (
../<name>/SKILL.md), so they stay siblings inskills/stack/and keep names unique in the repository. A change to a handoff (what a skill returns, or where it stops) updates the skill on both sides. - Unity skills link the
unity-coreskill and each other by relative path (../<name>/SKILL.md), so they stay siblings inskills/game/unity/. - When adding, renaming, moving, or removing a skill, update together its folder, its section's group in
.claude-plugin/marketplace.json, and its row in the matching section of theREADME.md"All skills" catalog. Edit "Choose by task" or "Requirements" only when the skill changes them. README.mdserves newcomers: choosing, installing, and prerequisites. Keep maintenance details out of it.
Applications under mcp/
- Keep each application's manifests, lockfile, dependencies, build, and tests inside its
mcp/<name>/workspace, with generated output, dependencies, and logs untracked. Install dependencies only when the task works on that application. - Tauri agent kit: publish only generic fixture evidence. Private application captures, logs, and integration backups stay outside the repository and release artifacts. Before native-input tests, reserve the Windows desktop and target only fixture processes you launched; a dispatch acknowledgement or a safe rejection does not prove an observed UI effect.
- Merging a version change to
mainpublishes the application through.github/workflows/<name>-publish.yml, so change a version only with explicit approval. After a failed run, use Re-run all jobs: each publication step publishes only what is missing.
Boundaries
- README installation commands (
npx skills add,npx skills update) are for end users. Do not run them from this repository: they change the agent's installed skills and check nothing here. - Testing or editing a skill does not authorize its side effects, such as image generation, nested Codex runs, or network installs.
Validation
| Changed area | Required checks |
| Skills and documentation | Affected claims, local links and anchors (excluding code examples), frontmatter, line counts, and folder, marketplace, and README consistency. Run git diff --check and review the diff, including untracked files. |
MCP server in mcp/<name>/ | From that folder, the step sequence of .github/workflows/<name>-check.yml, with the prerequisites from its README or CONTRIBUTING.md. Run steps that download tools only with network authorization. Preserve regression assertions rather than weaken them to pass. |
| Other helper scripts | Safe checks exposed by the owning skill, after confirming inputs and effects. |
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.