agent-extension-standards
Skill7 starsChanged 52 days agoMITView on GitHub
Current attested Codex, Claude Code, and Agent Skills extension contracts for skilltap. Use when designing, implementing, reviewing, or documenting harness adapters; plugin or marketplace manifests; native plugin lifecycle; skill directories or SKILL.md frontmatter; AGENTS.md/CLAUDE.md bridges; configuration scopes; compatibility transfer; adoption, reconciliation, status, or update behavior.
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 2 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
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 23
Use attestations under .research/attestation/ for source-direct verification.
- Very specific: line 24
Treat specialist briefs and the parent synthesis as analytical guidance, not primary sources.
- Very specific: line 29
Treat a skill as its complete directory with top-level SKILL.md.
- Very specific: line 31
Use .agents/skills/ as skilltap's canonical portable placement when the skill is compatible.
- Distinctive: line 33
Keep Codex and Claude plugin manifests, catalogs, identities, caches, scopes, consent, and update semantics distinct.
- Distinctive: line 35
Prefer deterministic native marketplace and plugin operations.
- Very specific: line 37
Keep ~/AGENTS.md as skilltap's canonical global instruction file and model ~/.codex/AGENTS.md and ~/.claude/CLAUDE.md as native bridges, …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill 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
Agent Extension Standards
Ground extension work in the current research substrate. Do not reuse remembered Claude-only marketplace assumptions or flatten the two harnesses into a generic plugin format.
Load the applicable contract
Read .research/analysis/briefs/current-agent-extension-standards.md before
making an architectural or behavioral decision in this domain.
Load a specialist brief when exact native details matter:
- Codex:
.research/analysis/campaigns/marketplace-standards/specialists/codex.md - Claude Code:
.research/analysis/campaigns/marketplace-standards/specialists/claude.md - Agent Skills:
.research/analysis/campaigns/marketplace-standards/specialists/agent-skills.md
Use attestations under .research/attestation/ for source-direct verification.
Treat specialist briefs and the parent synthesis as analytical guidance, not
primary sources. Never cite .research/.import-holding/ as evidence.
Preserve these boundaries
- Treat a skill as its complete directory with top-level
SKILL.md. Never manage only the Markdown file when sibling resources exist. - Use
.agents/skills/as skilltap's canonical portable placement when the skill is compatible. Create harness-native links or copies through adapters. - Keep Codex and Claude plugin manifests, catalogs, identities, caches, scopes, consent, and update semantics distinct.
- Prefer deterministic native marketplace and plugin operations. Do not mutate a native plugin cache as an installation API.
- Keep
~/AGENTS.mdas skilltap's canonical global instruction file and model~/.codex/AGENTS.mdand~/.claude/CLAUDE.mdas native bridges, including their precedence and divergence hazards. - Separate desired skilltap state, native declared state, and effective installed state.
- Classify transfer as faithful, materializable, partial, or blocked. Surface material consequences for piecewise operation-scoped approval.
- Register only sources selected by the user. Do not add search, browsing, ranking, recommendation, or inventory-discovery behavior.
Check freshness
The authoritative brief is dated 2026-07-10 and carries explicit revisit
triggers. When a decision depends on a native command, schema, path, or lifecycle
that may have changed, verify the current official primary source. If the
contract moved, refresh the .research/ attestations and synthesis before
changing foundation docs or implementation.
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.

