history-to-skilllisted

# History-to-Skill Turn a repository's accumulated engineering decisions into a reusable maintenance skill. Extract operating knowledge, not a generic repository summary. The finished skill should help an agent make changes that fit the codebase: use the right commands, respect component boundaries, follow review conventions, recognize recurring failures, and cite the evidence behind each claim. ## Modes Choose the smallest mode that satisfies the request: 1. **Full conversion**: collect evidence, synthesize a skill, validate it, and install it. 2. **Analyze only**: collect evidence and report the strongest patterns without writing a skill. 3. **Refresh**: collect newer history and revise an existing generated skill without erasing still-supported knowledge. ## Inputs Require one of: - A GitHub repository as `owner/repo` or a `github.com` URL. - A local git checkout. Accept an optional generated skill name. Default to `<repository-name>-maintainer`. Never request a token for a public repository unless the unauthenticated GitHub rate limit is exhausted. For private repositories, use an already-configured `GITHUB_TOKEN` or `GH_TOKEN`; never print or write the token. ## Step 1 — Locate the collector Prefer the `collect_repository_evidence` MCP tool when it is available. It gives the host agent a structured result without parsing terminal output. Otherwise, prefer the installed `history-to-skill` command. If it is on `PATH`, continue to Step 2. For a repository-only