Anthropic Skills
Anthropic Skills is Anthropic’s repository of task instructions, scripts and supporting resources for Claude. It is useful when designing reusable skills or examining concrete document-processing implementations. We pinned commit 2c5958ecac7cee2b583967a2c4981d0f4a75c579 and exercised its unmodified scripts on disposable inputs using CPython 3.12.13, pypdf 7.0.0, ReportLab 5.1.0 and PyYAML 6.0.3 on macOS 26.6.2. This was script execution, without skill installation in a Claude host or model calls.
All 19 skill references in five marketplace plugins existed. Eighteen of 19 root skills passed the repository’s minimal validator; claude-api exceeded its description-length limit. Artificial invalid naming and unknown metadata failed, while blank name and description passed. A generated .skill archive filtered selected caches and root evals, but included a fake .env file. PDF text and checkbox updates survived reopening with matching field/widget values and a legible rendering. Unknown fields, wrong pages and invalid values were rejected.
A legitimate dropdown value was also rejected: the selected extractor treated complete string options as their first characters. Direct pypdf filling of the same input accepted the full value. Treat this collection as inspectable implementation material, with task-specific input and output checks. Eighteen folders contained Apache-2.0 license files, while doc-coauthoring lacked a same-directory license. Host triggering, model effectiveness, OCR, Office conversion and cross-client behavior were not tested.
Read the full review, verification scope and sources →Features
Why choose it
Trade-offs
Maintenance check
- Last commit
- 2026-11-06
- Commits, last 90 days
- 15
- Releases, last 90 days
- 0
- Latest release
- —
- Open issues
- 1492
- Recent issues with at least one comment
- 3/13
Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.
Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.
Most discussed open issues
- #492 Security: Community skills distributed under anthropic/ namespace enable trust boundary abuse46 comments · 2026-03-02
- #228 Enable org-wide skill sharing in Claude.ai17 comments · 2026-01-13
- #556 run_eval.py: claude -p never triggers skills/commands (0% trigger rate across all queries)13 comments · 2026-03-07
- #1329 Proposing a second skill: compact-memory (symbolic notation for compact agent state)10 comments · 2026-06-17
- #62 All my skills have disappeared an now i get errors10 comments · 2025-10-24
Issue titles are quoted from GitHub. Open the link to check whether a problem still applies to the version you use.
Who it fits
- You want inspectable examples for a small, version-pinned set of tasks.
- You can validate produced files and inspect package members before sharing.
- You need proven model success rates or automatic cross-host compatibility.
- You expect the minimal validator or packager to certify correctness and safety.
Compatibility
Quick start
Use cases
Alternatives
Comments
No comments yet. Be the first!