What problem does it solve? Clipping a GitHub repository for a knowledge wiki is not like clipping a web page: a repo is hundreds or thousands of independently citable files, and dumping the whole tree either produces one unquotable blob or floods the vault with thousands of orphaned files. This skill clones a repo to a temporary directory, filters out dependency, build, binary, and lockfile noise, and produces a bounded set of Markdown clippings with per-file GitHub blob-URL citations. ## Core Features & Use Cases - Per-file mode for small repos: Repositories with up to 300 includable files become one clipping per file under raw/clippings/gh/<owner>/<repo>/, plus a _repo-overview.md manifest, with citations pointing to the exact blob URL at the clipped commit. - Digest mode for large repos: Repos over the cap produce a bounded manifest with composition and module tables, up to 150 module listing documents (paths and sizes only, never content), and up to 15 full-content anchor clips such as README, LICENSE, and project manifests. - Content-hash dedup and pruning: Re-clips skip unchanged files via source-hash comparison, and digest-mode re-runs automatically prune stale listings when module grouping changes. - Use Case: Capture a 10,000-file enterprise monorepo as 150 bounded module listings plus anchor files, then hand off to wiki-ingest to write a small set of curated source summaries citing the digest as evidence. ## Quick Start Ask the agent to clip a GitHub repository into the wiki by running the clip-gh script with the owner/repo, a quality tier, and the research topic, for example: clip the repository owner/repo at high quality under the topic of my current research run.