santiagosainz-skills
AGENTS.md1 starsChanged 3 months agoMITView on GitHub
Portable catalog of maintainer-workflow skills for codex/opencode/pi — not an application. Definition of done: make check passes. Real install (writes to ~/.codex/skills, ~/.agents/skills, ~/.pi/agent/skills): scripts/validate-catalog.sh requires jq. It checks: catalog.json is valid JSON; every skill entry has non-empty name, group, status, path, provenance, runtimeBehavior; targets is a non-empty array drawn only from codex/opencode/pi; name matches ^[a-z0-9]+(-[a-z0-9]+)*$; the skill's path directory exists; and that directory contains a SKILL.md. scripts/publish-public.sh [owner] creates and pushes the public GitHub repo via gh — requires…
- Deletes or force-pushes
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 32,671 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 repository9 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- architecture-reviewSkill
- debugSkill
- domain-modelingSkill
- handoffSkill
- implementation-planSkill
- requirements-reviewSkill
- test-and-verifySkill
- webgl-scroll-siteSkill
The file
We split this AGENTS.md into the rules it gives an agent
6 rules. About 0 are common, 0 are distinctive, and 6 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.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 65
Create skills/<group>/<name>/SKILL.md with frontmatter name + description and the workflow body (contract, use cases, numbered steps), …
- Very specific: line 66
Add skills/<group>/<name>/agents/openai.yaml with a matching name, short description, and a prompt invoking the skill.
- Very specific: line 67
Add a corresponding entry to catalog.json's skills[] array: name, group, status, path: "skills/<group>/<name>", targets (which runtimes), …
- Very specific: line 68
If a runtime needs special handling notes, update the relevant adapters/<runtime>/README.md.
- Very specific: line 69
Run make check (or at minimum scripts/validate-catalog.sh) to confirm the catalog entry and SKILL.md are well-formed.
- Very specific: line 70
Run scripts/install.sh --dry-run to confirm the new skill's copy plan targets the intended runtime skill roots.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 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
santiagosainz-skills Agent Notes
Portable catalog of maintainer-workflow skills for codex/opencode/pi — not an
application. Definition of done: make check passes.
Commands
# Full check: regenerate assets, validate catalog, dry-run install
make check
# Validate catalog.json + skill layout only
make validate
# equivalent: bash scripts/validate-catalog.sh
# Dry-run the installer (prints copy plan, writes nothing)
make dry-run
# equivalent: bash scripts/install.sh --dry-run
# Default target (validate + dry-run, no asset regen)
make
# equivalent: make all
# Regenerate Random Access-themed SVG assets (wordmark, workflow-spine, social preview)
make assets
# equivalent: python3 scripts/render_assets.pyReal install (writes to ~/.codex/skills, ~/.agents/skills, ~/.pi/agent/skills):
scripts/install.sh # all targets, all skills
scripts/install.sh --target codex # one runtime
scripts/install.sh --target opencode
scripts/install.sh --target pi
scripts/install.sh --core-only # skip non-core groupsscripts/validate-catalog.sh requires jq. It checks: catalog.json is valid JSON; every skill entry has non-empty name, group, status, path, provenance, runtimeBehavior; targets is a non-empty array drawn only from codex/opencode/pi; name matches ^[a-z0-9]+(-[a-z0-9]+)*$; the skill's path directory exists; and that directory contains a SKILL.md.
scripts/publish-public.sh [owner] creates and pushes the public GitHub repo via gh — requires a clean working tree and runs validate-catalog.sh first. Not a routine dev command.
CI (.github/workflows/validate.yml) runs exactly scripts/validate-catalog.sh and scripts/install.sh --dry-run on every PR and push to main — same as make validate dry-run. A second workflow, .github/workflows/skill-review.yml, dogfoods this repo's own requirements-review and architecture-review skills against each PR diff via opencode run and posts the result as a PR comment (read-only tool permissions; see docs/codex-automation.md).
There is no test suite beyond catalog/install validation — skills are markdown instructions, not executable code.
Architecture
This repo is a portable catalog of maintainer-workflow skills, not an application. Four pieces:
skills/<group>/<skill-name>/— the canonical, runtime-agnostic definition of each skill. Two groups today:core(seven maintainer-workflow skills) anddesign(webgl-scroll-site). Each skill directory holdsSKILL.md(YAML frontmatter withname+description, then the actual workflow instructions: contract, when to use it, numbered steps) andagents/openai.yaml(a short Codex/agents-SDK-style wrapper:name, one-linedescription, and apromptthat tells the agent to "use the<skill>skill" and restates the gist).SKILL.mdis the source of truth;agents/openai.yamlis a thin pointer at it for tooling that expects that shape.catalog.json— the index. Top-levelinstallDestinationsmaps each runtime to its skill root (codex→~/.codex/skills,opencode→~/.agents/skills,pi→~/.pi/agent/skills). Each entry inskills[]declaresname,group/status(e.g.coreordesign),path(pointing intoskills/<group>/...),targets(which ofcodex/opencode/piget it),provenance, andruntimeBehavior(a one-line note on how the skill's behavior differs per runtime). This is the only place that decides which skills ship to which runtimes — skill directories themselves are runtime-agnostic.adapters/<runtime>/README.md— per-runtime usage notes (codex, opencode, pi), not code. Each documents how that runtime should track state and verify work while running a skill (Codex: conversation/tool checklist, shell-first; OpenCode: nativetodowrite+/plan//verify; Pi: file-backedPLAN.md/STATE.md/VERIFY.md/TODO.md) and restates that runtime's install destination. Adapters do not contain skill logic — they are operating notes for whoever (human or agent) is running skills under that tool.scripts/— the only executable layer:validate-catalog.shenforces the catalog schema described above and that every referenced path/SKILL.mdexists.install.shreadscatalog.jsonwithjq, filters by--targetand--core-only, and copies each selected skill'spathdirectory into the destination for each target runtime (skip-if-unchanged viadiff -qr, otherwiserm -rf+cp -R).--dry-runonly prints the copy plan.render_assets.pyregenerates the SVGs inassets/from the Random Access theme tokens — unrelated to skill packaging.publish-public.shis a one-time/occasional repo-publish helper (validate →gh repo create→ push), not part of the authoring loop.
Authoring workflow — adding a new skill:
- Create
skills/<group>/<name>/SKILL.mdwith frontmattername+descriptionand the workflow body (contract, use cases, numbered steps), following the existing skills as the template. Workflow skills go incore; design skills go indesign. - Add
skills/<group>/<name>/agents/openai.yamlwith a matchingname, shortdescription, and apromptinvoking the skill. - Add a corresponding entry to
catalog.json'sskills[]array:name,group,status,path: "skills/<group>/<name>",targets(which runtimes),provenance,runtimeBehavior. - If a runtime needs special handling notes, update the relevant
adapters/<runtime>/README.md. - Run
make check(or at minimumscripts/validate-catalog.sh) to confirm the catalog entry andSKILL.mdare well-formed. - Run
scripts/install.sh --dry-runto confirm the new skill's copy plan targets the intended runtime skill roots; runscripts/install.sh(optionally--target <runtime>) to actually surface it under~/.codex/skills,~/.agents/skills, or~/.pi/agent/skills.
CI re-runs steps 5–6 on every PR, and a second workflow exercises two of the published skills directly (requirements-review, architecture-review) against the PR diff via OpenCode as a live correctness check.
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.

