metaskills
AGENTS.md1 starsChanged 14 days agoMITView on GitHub
MetaSkills version: 1.1.0 MetaSkills owns the user's canonical shared skills, command wrappers, and agent roles, installed into Claude Code, opencode, and Codex user-level configs. Canonical skills are project-agnostic: resolve facts from the consuming project's AGENTS.md under AGENTS.template.md; stop rather than guess missing facts. Never embed repo slugs, board IDs, branch defaults, environment URLs, cluster/image names, credentials, deployment facts, or local-machine-only paths. Mostly Markdown, TOML, PowerShell, and Bash, plus the paper-doc browser assets and Python scaffold. Keep changes small, explicit, and…
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 31,439 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 repository15 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- backlog-refineSkill
- burndownSkill
- code-reviewSkill
- dev-cycleSkill
- issue-diagnoseSkill
- issue-implementSkill
- issue-planSkill
- issue-raiseSkill
- issue-refineSkill
- paper-docSkill
- project-statusSkill
- qaSkill
- setupSkill
- weekly-reviewSkill
The file
We split this AGENTS.md into the rules it gives an agent
6 rules. About 1 is common, 4 are distinctive, and 1 is very specific to this project.
≈1
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 24
Update affected surfaces together:
- Distinctive: line 34
No compiled application. Run sync lint first, then inspect and dry-run:
- Distinctive: line 63
Read changed skills/agents to verify the project-fact contract and layout invariants above.
- Distinctive: line 69
No board documented. Skills needing board owner, project number, or Status option names must stop and ask.
- Very specific: line 71
If added, record owner/org, project number, and Status lifecycle here.
- Common: line 89
Not applicable. This repository has no running web application, browser login flow, test credentials, or storage-state file.
≈ “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 - MetaSkills
MetaSkills version: 1.1.0
Project Overview
MetaSkills owns the user's canonical shared skills, command wrappers, and agent roles, installed into Claude Code, opencode, and Codex user-level configs.
Canonical skills are project-agnostic: resolve facts from the consuming project's AGENTS.md under AGENTS.template.md; stop rather than guess missing facts. Never embed repo slugs, board IDs, branch defaults, environment URLs, cluster/image names, credentials, deployment facts, or local-machine-only paths.
Code Layout & Tech Stack
skills/<name>/SKILL.md: canonical skills shared across all three harnesses.commands/<name>.md: thin same-named skill +$ARGUMENTShandoffs for opencode commands and Codex prompts, never duplicated workflows. Claude Code uses skills directly as slash commands; no wrapper copies.agents/opencode/*.md,agents/claude/*.md,agents/codex/*.toml: subagent/role definitions with identical bodies across dialects; configuration syntax differs.AGENTS.template.md: required consuming-project contract, including the MetaSkills version and its change list. Bump minor when projects must update theirAGENTS.mdor issue templates, patch for low-impact changes, major for pride releases.sync.ps1,sync.sh: managed harness installers..opencode/skills/: OpenCode-only repository maintenance. Mirror intoskills/only if behavior should install globally.skills/paper-doc/: portable document-generation skill with HTML/CSS/JavaScript assets, a native SVG C4 viewer, a Python scaffold, and authoring references. Its paired wrapper iscommands/paper-doc.md.docs/: GitHub Pages site, published frommaster/docs.index.htmlis a self-contained paper-doc workflow guide; each diagram is inline SVG rendered from the PlantUML or draw.io source embedded beside it. It is pinned to the commit in its masthead. Never installed.
Mostly Markdown, TOML, PowerShell, and Bash, plus the paper-doc browser assets and Python scaffold. Keep changes small, explicit, and portable across Windows/Unix.
Update affected surfaces together:
- Skills: canonical
skills/<name>/SKILL.md; commands only when handoffs change. - Roles: all three dialects.
- Installation: both sync scripts, including added/renamed/removed managed files.
- Layout, installation, or workflows:
README.md. - Anything projects must change:
AGENTS.template.mdversion bump and change entry, plus this file's version line.
Build & Validation
No compiled application. Run sync lint first, then inspect and dry-run:
git status --short
git diff --check
.\sync.ps1 -Check
.\sync.ps1 -WhatIfFor Unix-shell changes, also review sync.sh and run when Bash is available:
./sync.sh --check
./sync.sh --dry-run--check / -Check verifies skill frontmatter (name, description), bidirectional skill/command pairing, three-dialect agent body parity, PowerShell-only shell samples, and em/en dashes. Both scripts must report identical findings.
Run python3 tests/test_installers.py on Linux (bare or container) for isolated installer regression tests. The suite is Bash-only; Windows and PowerShell coverage were removed, and it skips entirely off Linux. Installer tests must cover hard-linked metadata as well as symlinks, and byte-sensitive fixtures must avoid platform newline translation. tests/workflow-scenarios.md contains manual instruction evaluations, not executable behavioral tests.
For paper-doc scaffold or viewer changes, run the dependency-free regressions (Python 3.10+, Node.js 18+):
python tests/test_paper_doc_scaffold.py
node --test tests/test_paper_doc_c4.cjsOpen skills/paper-doc/assets/template/index.html directly using a file URL for browser verification. Check relevant interactions, desktop/mobile layout, and console errors when assets change; use skills/paper-doc/references/c4.md for the diagram contract. Generated documents must remain compatible with direct file opening. Both sync scripts discover the skill directory and paired command automatically and copy all bundled assets.
Read changed skills/agents to verify the project-fact contract and layout invariants above; lint alone does not validate behavior.
DB tripwire files: none. This repository has no database layer or integration database suite.
Project Board
No board documented. Skills needing board owner, project number, or Status option names must stop and ask.
If added, record owner/org, project number, and Status lifecycle here. Look up field/option IDs live with gh project field-list; never record them.
Repositories
- App repo slug:
odytrice/metaskills. - Deployment repo: none.
- Issue template path: none documented. If no template exists, issue workflows should use their fallback headings.
Environments
None. This repository has no running environments.
Branch Map
master: primary integration branch and PR base unless the user explicitly names another base.
Agent Login
Not applicable. This repository has no running web application, browser login flow, test credentials, or storage-state file.
Review Notes
Prioritize instruction drift and cross-harness parity over style. Enforce the project-fact, wrapper, role, and installer invariants above. Never alter user-level configs except through sync scripts or an explicit user request.
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.

