Pro Skill Packs logo PRO SKILL PACKS

Study / How portable are the most-installed agent skills?

How portable are the most-installed agent skills?

We scanned the most-installed agent skills on skills.sh for ties to one assistant: tool names, paths, frontmatter keys and scripts. Data and scripts included.

Pro Skill Packs, 2026-10-04. Every number below comes from data.csv (one row per skill) and summary.json, both produced by the scripts in this folder. No model judged anything: each measure is a regular expression or a file listing.

Method

  1. Took the top 100 skills on skills.sh by all-time installs (the leaderboard embedded in skills.sh, 2026-10-04). Install counts there are per package, so skills from the same publisher often share nearly the same number.
  2. For each, found SKILL.md through one GitHub tree listing per repo (fetch.py), then fetched the raw file once, 0.5 seconds apart. Skills whose source is not a GitHub owner/repo, or whose folder was not found, were skipped.
  3. analyze.py measured: frontmatter keys beyond name and description; vendor tool names (WebFetch, Skill tool, AskUserQuestion, Bash(...), mcp__, and similar); agent product names (Claude Code, Codex, Cursor, Gemini CLI, Copilot, CLAUDE.md or GEMINI.md); hard-coded vendor paths (~/.claude/, .cursor/, .codex/, .gemini/, $CLAUDE_*); the generic .agents/ path; bundled scripts by file extension; length in lines; and a phrase test for a fallback ("if you cannot...", "ask the user to paste", "otherwise ask").
  4. "Coupled" means at least one of: a vendor tool name, an agent product name, a vendor path, allowed-tools, or a frontmatter key outside the public spec. The spec (agentskills.io/specification) lists name, description, license, compatibility, metadata and allowed-tools.

Sample

  • 100 listed. 69 fetched, from 14 repos. 27 skipped (source is not a GitHub repo), 4 skipped (skill folder not found in the repo).
  • Three publishers supply 47 of the 69, so the skill-level counts lean toward their habits. Repo-level numbers are given where it matters.
  • The fetched skills hold 48,930,288 of the install counts in the sample; coupled skills hold 20,992,884 of them (43%). Install counts are inflated by bundles, so read that as weight, not as users.

Findings

  1. Most are portable by these tests: 40 of 69 (58%) show no coupling. 29 (42%) show at least one. At repo level, 10 of 14 repos have at least one coupled skill and 4 are clean throughout.
  2. Hard-coded vendor paths: 0 of 69. No skill points at ~/.claude/, .cursor/, .codex/ or .gemini/, and none uses a $CLAUDE_* variable. The generic .agents/ path also appears in 0. The path problem people worry about is not in this sample.
  3. Frontmatter is where the coupling is. 18 of 69 use only name and description. 51 add keys, mostly spec keys (metadata 31, license 21). 17 use a key outside the spec: disable-model-invocation 12, version 3, argument-hint 2, and one each of hidden, displayName, emoji, homepage. 5 use allowed-tools, each naming a command-line tool through Bash(...).
  4. Vendor tool names in the text: 17 of 69. By tool (a skill can name several): Skill tool 7, Bash(...) 5, WebFetch 3, AskUserQuestion 1, mcp__ 1. Excluding Bash(...), which only appears in allowed-tools, it is 12 of 69.
  5. Agent product names: 6 of 69 (Claude Code 4, Codex 4, Cursor 2, CLAUDE.md or GEMINI.md 3). None names Gemini CLI or Copilot. A mention is not always lock-in: some list products the skill works with.
  6. Fallbacks are rare: 17 of 69 contain any fallback phrase. Among the 29 coupled skills, 23 have none, so when the named tool is missing the instructions stop. (Phrase test only; see limits.)
  7. Scripts: 14 of 69 bundle code, 55 are plain text. Languages among the 14 (a skill can have several): shell 7, PowerShell 7, JavaScript 5, Python 4, TypeScript 1. Only 4 of the 14 also contain a fallback phrase. Skills that need a runtime are portable only where that runtime exists.
  8. They are short. Median 105 lines; 56 of 69 are 200 lines or fewer; 1 is over 500 (564 lines). 35 of 69 have a references/ folder for long material. No description exceeds the 1,024-character limit (median 375).

What ties a skill to one agent

Bundled scripts by language

Limits

  • The top 100 is skewed: 27 listed skills have a source that is not a GitHub repo, so they are missing, and three publishers make up 47 of the 69 fetched.
  • Regexes see strings, not meaning. "Works with Claude Code, Cursor and Codex" counts as a product mention but is the opposite of lock-in. A fallback phrase does not prove the fallback works.
  • Only SKILL.md text was scanned. Coupling inside references/ files or scripts was not measured.
  • A skill can need nothing vendor-specific and still behave differently between assistants. We did not run any of them.
  • One snapshot, one day. Skills change, and install counts move.

How to make a skill portable

  1. Keep frontmatter to name and description. Add license, compatibility or metadata if you want. If a tool-specific key helps in one product, make sure the skill still works when another tool ignores it.
  2. Name the capability, not the tool: "fetch the page with whatever web tool you have", not "use WebFetch".
  3. For every capability, say what to do without it: no web access, ask the user to paste the text; no code execution, say the counts are unverified.
  4. Write the stop condition first. Before you write a fallback, write what clears it: the evidence that the missing tool or data is back, or that the user's pasted text is enough. Otherwise a fallback becomes the permanent route. (Suggested by an agent on Moltbook.) Add a "resume when:" line next to each stop: say which condition lets the skill continue, for example "resume when: the user pastes the page text, or the tool is back". (Also suggested by an agent on Moltbook.)
  5. Use paths relative to the skill folder. In install notes, give .agents/skills/ first and tool-specific folders second.
  6. Scripts: pick one common language, stick to its standard library, state the runtime in the README, and avoid shell-only or PowerShell-only logic for anything essential.
  7. Write the description to say what the skill does and when to use it, under 1,024 characters.
  8. Keep SKILL.md short. Move long reference material into files the skill links to.
  9. Ship a paste-in prompt version for chat apps that cannot load skills.
  10. Test by reading the skill as an assistant with no special tools. If any step is impossible, add the fallback.

Reproduce

python3 fetch.py && python3 analyze.py && python3 charts.py (needs curl and Python 3; GitHub's unauthenticated API allows enough calls for this sample). Data: data.csv, summary.json, fetch_log.csv.

We make portable skills, with a paste-in prompt for every one. The free ones are at https://github.com/proskillpacks/skills

Check your own skills

We turned these checks into a free tool: Skill Portability Check.

Files

data.csv summary.json fetch_log.csv fetch.py analyze.py charts.py