osanwe-public / skills
AGENTS.md0 starsChanged 52 days agoMITView on GitHub
Every harness reads this file on its walk into .agents/skills/ (Codex/OpenCode/Goose/Crush/Cline/Pi natively; Claude Code via the .claude/skills/AGENTS.md pointer). This is the skill-authoring rulebook; author rules HERE.
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.
The file
We split this AGENTS.md into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is 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.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 7
Before building or materially extending any skill: /gate b (root invariant.
≈ “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
Skill authoring (binds work under .agents/skills/ -- CANON tree) -- canonical
Every harness reads this file on its walk into
.agents/skills/(Codex/OpenCode/Goose/Crush/Cline/Pi natively; Claude Code via the.claude/skills/AGENTS.mdpointer). This is the skill-authoring rulebook; author rules HERE.
.agents/skills/<name>/is the CANONICAL home of every skill (SKILL.md + ref-*.md + scripts/ + assets/)..claude/skills/<name>/is a GENERATED derived copy (.agents/scripts/sync.py); NEVER hand-edit a derived file -- edit canon, re-run sync, commit both together. Consistency:python .agents/scripts/checkall.py.- MIGRATION WINDOW (S3, 2026-08): a canon SKILL.md still carrying the legacy "CODEX ADAPTER" marker is an unmigrated stub -- for those skills the live
.claude/skills/<name>/remains canonical until their batch lands; sync/driftcheck skip them. - Before building or materially extending any skill:
/gate b(root invariant; sheet + registry row via tools/gate-eval.py). - Canon frontmatter is SPEC-ONLY (agentskills.io): top-level
name(== dirname, 1-64 lowercase-hyphen, no '--') +description(double-quoted, ASCII, <= 500 chars hard cap, median target ~300, trigger keywords in the first 200 chars, any Distinct-from disambiguation preserved in substance) +metadata(string->string map carrying every Osanwe field stringified: osanwe-risk, osanwe-effort, osanwe-arguments, osanwe-argument-hint, osanwe-allowed-tools space-separated, osanwe-categories, osanwe-type, osanwe-status, osanwe-created, osanwe-updated). NO top-level allowed-tools in canon -- sync re-materializes it in the Claude derived copy from metadata (no privilege widening). - SKILL.md body: lean main file, aspirational <=263 lines; detail in one-level-deep
ref-*.md. Legitimate multi-phase skills keep full content inline -- vault-audit X2 flags HARD DRIFT only above 900 lines. Quality preservation beats line-count extraction. - Subagent-bearing phases use harness-conditional dispatch phrasing: dispatch the named subagent when an Agent tool exists; otherwise execute the documented inline fallback (every such skill carries it verbatim). No skill conditions behavior on model identity. DW-topology skills: sequential spine is the universal fallback; fan-out concurrency <=6; budgets via
<SKILL>_DW_TOKEN_BUDGETenv vars. - Shared-discipline primitives (F11/F14/F17, F.halt, Pattern 6, Pre-Output HALT gate, Phase J.0/O.0 pre-commit audit gate, FOLLOWUPS:skills): Vault Codex Section 4 glossary; canonical F11 lifecycle: Atlas/sources/meta/ref-execution-discipline.md.
- Registry of record: Vault Codex Section 4 (+ changelog for post-2026-07-06 changes). Retire skills to
.claude/skills/_archive/(never-touch for sync), never delete. - Done-bars for a new/edited canon skill: strict YAML parse;
python skeleton/eval/eval-skill-overlay.py --checkexit 0 (canon-spec schema);agentskills validate(PyPI skills-ref) pass;python .agents/scripts/checkall.pygreen.
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.