skill-portability-audit
Skill0 starsChanged 2 months agoMITView on GitHub
Audit whether an agent skill works across every harness (Claude Code, DeepSeek Harness dsh, Codex, Gemini CLI, Copilot, Cursor) before publishing or after a "my skill loads in Claude but not in dsh" report. Checks name collisions across roots, tier shadowing, discovery rules, and trigger quality per agent.
- Reads credentials
In this repository8 other files it gives its agents.
- AGENTS.mdAGENTS.md
- improve-claude-mdSkill
- show-meSkill
- claude-plugin-devSkill
- deepseek-cost-routerSkill
- deepseek-migrationSkill
- dsh-plugin-devSkill
- skill-authorSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is 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.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 24
Nested SKILL.md: Every harness scans only the *direct children* of a skill root.
- Distinctive: line 41
Run these in order; stop and fix at the first red flag.
- Distinctive: line 79
When multiple issues exist, fix in this order — discovery problems first (invisible skills), then trigger problems (loads but never fires), …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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
Auditing a skill for cross-agent portability
A skill that works in one agent frequently breaks silently in another. The failure modes are specific and checkable — run this audit before publishing, and whenever a user reports "it loads in X but not in Y".
The five silent failure modes
- Name collision + tier shadowing. dsh and Codex read both
<project>/.agents/skills(tier 200) and~/.agents/skills(tier 500). The same skill name at both levels: the project copy wins silently — no warning, ever. If a user's project has an unrelated skill with your name, yours never loads. - Nested SKILL.md. Every harness scans only the direct children of a
skill root.
skills/a/SKILL.mdloads;skills/a/b/SKILL.mdis invisible in all of them. Monorepos that nest skills under product directories hit this constantly. - Name mangling. Frontmatter
name: My Skill!gets normalized differently by different tools — some mangle, some reject. The safe form is kebab-case matching the directory name exactly. - Trigger starvation or flooding. The
descriptionis the only thing agents match requests against. "A collection of helpers" never triggers; a description listing every keyword triggers on everything and dilutes routing. - Format assumptions. Flat
my-skill.mdfiles are a dsh convenience — Claude Code only reads<dir>/SKILL.md. And relative links to files outside the skill folder break the moment the skill is copied.
The audit procedure
Run these in order; stop and fix at the first red flag.
# 1. Structural lint (frontmatter, nesting, links, orphans)
aipx lint skills/
# 2. Where does each agent actually see it?
aipx list # user-scope view, per agent root
aipx list --project # project-scope view
# 3. Collision check across ALL roots — the shadowing trap:
# does the same skill name exist in more than one root?
# (compare the output of `aipx list` against the project roots)Manual checks the tooling can't do for you:
nameis kebab-case AND equals the directory namedescriptionanswers "when should I fire?" in one sentence — read it and ask: would an agent route a real user request here? Too broad? Too vague?- SKILL.md body ≤ ~500 lines; deep material moved to
references/ - No skill references files outside its own folder
- If a project root also has this skill name: renamed or intentional?
Cross-agent smoke matrix
Install once with aipx, then verify in each agent the user cares about:
| Agent | Verify how |
| Claude Code | skill appears in /doctor output; new session needed |
| dsh | / command palette → Skills group; needs DEEPSEEK_API_KEY |
| Codex | appears in skill listing on next run |
| Gemini CLI | next session; check aipx list root exists |
Fix priority
When multiple issues exist, fix in this order — discovery problems first (invisible skills), then trigger problems (loads but never fires), then cosmetics:
- naming/collision/shadowing → 2. description trigger → 3. size/structure →
- links → 5. style.
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.

