What problem does it solve? Maintainers of watched GitHub repositories often struggle to decide what to work on next, defaulting to vague ideas like "improve docs" or "clean up code". This Skill produces five concrete, implementable action ideas anchored to the repo's real current state — open issues, stale PRs, TODO comments, missing CI files, or outdated dependencies — each scored and priority-ranked with a Top Pick verdict. ## Core Features & Use Cases - Anchored idea generation: Every idea must cite a real anchor such as an open issue number, a grep-matched TODO, a missing LICENSE or CI workflow, or a named dependency at a known version. - Four quality gates: Specificity linting rejects filler phrases, novelty checks prevent repeating ideas from the last 14 days, an implementability gate demotes items needing human decisions to a Monitor section, and a leverage/concreteness/novelty score ranks survivors. - Single-call state fetch: Pulls repo metadata, README, manifest files, issues, and PRs via one GitHub GraphQL query, with retry and degraded-mode fallbacks. - Use Case: A developer watching several repos runs the skill each morning and receives an article at articles/repo-actions-<date>.md listing five ranked actions, a Top Pick for tomorrow, and a notification summary. ## Quick Start Ask the agent to run the repo-actions skill to generate today's five prioritized action ideas for the watched repository configured in memory/watched-repos.md.
