cabral-skills
AGENTS.md1 starsChanged 4 months agoApache-2.0View on GitHub
How to work on this repository. Canonical contributor guide; CLAUDE.md symlinks to it. My personal source of truth for two artifact kinds, both activated with native mechanisms (no external orchestrator): - Skills (skills/<name>/) — standalone, npx-installable Agent Skills. Self-contained. - Plugins (plugins/<name>/) — self-contained Antigravity plugins installed with agy plugin install. A plugin bundles everything it needs inside its own dir: skills/, optional agents/, hooks.json, mcp_config.json, commands/, scripts/, rules/, and a plugin.json. There is one real plugin today — adk-developer.…
- Commits and 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 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 repository5 other files it gives its agents.
The file
We split this AGENTS.md into the rules it gives an agent
9 rules. About 0 are common, 4 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 21
When you add or change a standalone skill under skills/<name>/:
- Very specific: line 23
Keep the standard layout: SKILL.md (with name + description frontmatter) at the root, plus optional references/, examples/, scripts/.
- Very specific: line 25
If it ships scripts, make them executable (chmod +x) and document how to run them in SKILL.md.
- Distinctive: line 27
Run the validation checklist before tagging.
- Distinctive: line 48
Write your own skills/rules directly under the plugin.
- Very specific: line 53
Add a per-plugin sync script under scripts/ (model it on scripts/vendor-agents-cli.sh): download the upstream at a pinned tag, copy the …
- Very specific: line 57
Add the upstream to the drift workflow matrix (.github/workflows/check-vendored-upstreams.yml) so a newer release opens a tracking issue.
- Distinctive: line 125
Run the validation checklist.
- Distinctive: line 137
Run the repository validation CLI to verify manifest parity, frontmatter, script permissions, workspace hygiene, link integrity, vendor …
≈ “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 — cabral-skills maintainer guide
How to work on this repository. Canonical contributor guide; CLAUDE.md symlinks to it.
What this repo is
My personal source of truth for two artifact kinds, both activated with native mechanisms (no external orchestrator):
- Skills (
skills/<name>/) — standalone,npx-installable Agent Skills. Self-contained. - Plugins (
plugins/<name>/) — self-contained Antigravity plugins installed withagy plugin install. A plugin bundles everything it needs inside its own dir:skills/, optionalagents/,hooks.json,mcp_config.json,commands/,scripts/,rules/, and aplugin.json.
There is one real plugin today — adk-developer. Retired scaffolding lives in archive/
(in-tree, inactive; see archive/README.md).
Working on a skill
When you add or change a standalone skill under skills/<name>/:
- Keep the standard layout:
SKILL.md(withname+descriptionfrontmatter) at the root, plus optionalreferences/,examples/,scripts/. - If it ships scripts, make them executable (
chmod +x) and document how to run them inSKILL.md. - Update the README skills table.
- Run the validation checklist before tagging.
Authoring a plugin (the blend recipe)
This is the reusable process — the pattern for any future plugin (e.g. an Apigee plugin), not just
adk-developer. A plugin is a self-contained directory that agy plugin install <dir> can
consume directly. Blend = your authored material + upstream material kept updated, all bundled:
plugins/<name>/
plugin.json # name, version, description, skills[] (manifest of bundled skills)
skills/ # BOTH authored and vendored skill bodies live here
<my-authored-skill>/
<vendored-upstream-skill>/
rules/*.md # authored rules (see "Rules" — installer ignores these; copied per project)
mcp_config.json # optional: MCP servers
hooks.json # optional: hooks (reference scripts by ABSOLUTE path — relative CWD fails)
agents/ commands/ # optionalAuthored parts
Write your own skills/rules directly under the plugin. They are the source of truth — edit in place.
Vendored parts (kept updated)
To blend in an external upstream you want to track:
- Add a per-plugin sync script under
scripts/(model it onscripts/vendor-agents-cli.sh): download the upstream at a pinned tag, copy the chosen skill dirs intoplugins/<name>/skills/, and write provenance (repo/tag/commit/license/skills) intovendored.json. - Add the upstream to the drift workflow matrix
(
.github/workflows/check-vendored-upstreams.yml) so a newer release opens a tracking issue. It never auto-syncs — syncing stays deliberate and pinned. - Never hand-edit vendored skill dirs — edits are lost on the next sync. Fix upstream (or fork and point the sync at your fork).
When to vendor vs point: if you just use an external plugin/skill as-is, point at it in
SOURCES.mdandagy plugin install/npx skills addit from upstream. Vendor only when you need to pin it or blend it into a plugin alongside your own material.
Rules are a bundled plugin component
Per the Antigravity plugin schema, rules/*.md are a first-class bundled component (always-on,
Priority 0). When a plugin is discovered as a workspace-local plugin under
.agents/plugins/<name>/, its rules load automatically along with skills/agents/hooks/MCP. So
author rules under plugins/<name>/rules/ — they travel with the plugin; no manual copy needed for
the workspace-local path.
Open caveat (see ROADMAP.md): whether a globally-installed plugin
(agy plugin install) also surfaces its bundled rules is still being verified. Until confirmed, if
a global install doesn't apply the rule, copy it into the project as a fallback:
mkdir -p .agents/rules && cp /path/to/plugins/<name>/rules/*.md .agents/rules/(Optionally ship a tiny apply-rules.sh in the plugin for that fallback.)
plugin.json
Carry name, version, description, and a skills[] array listing the bundled skill dir names
(a manifest of what's inside skills/). Bump version on every change.
Vendored upstreams
Tracked in vendored.json; a scheduled Action
(.github/workflows/check-vendored-upstreams.yml) watches each and files an issue on a newer
release (never auto-syncs).
google/agents-cli → plugins/adk-developer/skills/
The google-agents-cli-{workflow,scaffold,adk-code,eval,deploy,publish,observability} skills are
vendored from google/agents-cli (Apache-2.0), pinned to an
upstream tag. (google-agents-cli-adk-frontend and google-agents-cli-adk-auth in the same dir are authored — not vendored,
safe to edit.)
To sync to a newer upstream release:
scripts/vendor-agents-cli.sh <upstream-tag> # e.g. v1.2.0This downloads the tag, replaces the seven vendored dirs under plugins/adk-developer/skills/, and
rewrites vendored.json. Then: review the diff, update the README if descriptions changed, run the
validation checklist, and cut a new release tag (a sync is at least a
minor bump).
Consuming this repo
- A standalone skill:
npx skills add carlosmscabral/cabral-skills --skill <name>(or omit--skillfor all five). Readsskills/only. - The ADK plugin: global via a remote subfolder URL
(
agy plugin install https://github.com/carlosmscabral/cabral-skills/tree/main/plugins/adk-developer) or workspace-local by copying/symlinking it into.agents/plugins/(auto-discovered). See README "Develop ADK in a project"; note the rules caveat above. - External stuff: see SOURCES.md.
Releasing
- Land changes on
mainwith the README updated. - Run the validation checklist.
- Cut a semver tag and push:
git tag -a vX.Y.Z -m "vX.Y.Z: <what changed>" git push origin main vX.Y.Z- patch — content fix within a skill/plugin.
- minor — new skill/plugin/capability, or a vendored-upstream sync.
- major — a breaking layout change (removed/relocated skills or plugins).
Validation checklist
Run the repository validation CLI to verify manifest parity, frontmatter, script permissions, workspace hygiene, link integrity, vendor digests, and trigger overlaps:
python3 scripts/validate_repo.pyTo auto-repair fixable issues (such as script permissions, vendor digests, or __pycache artifacts):
python3 scripts/validate_repo.py --fixDoes 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.

