What problem does it solve? After a pull request merges, there is no structured way to capture how good its lede (summary) actually was. This Skill records the author's rating of a merged PR's lede into the codeassembly event corpus, building a feedback dataset that shows what good ledes look like and where drafting guidance fails. ## Core Features & Use Cases - Guided rating capture: Presents the agent-written lede and the merged lede side by side, then asks the author for a five-level rating (poor through exemplary) with an optional free-text comment. - Automatic verdict derivation: Derives whether the lede was accepted or revised by comparing the two texts, and fingerprints the doctrine (entry-drafter and lede-cutter bodies) in force when the draft was written via a doctrine-hash. - Corpus integrity safeguards: Writes only author-given ratings, refuses unreachable or readonly stores, and supports PRs merged outside the normal merge flow via explicit lede files. - Use Case: After merging PR #142, invoke the skill to inspect the episode, review both ledes, rate the merged lede as "strong" with a comment, and write one tagged event record to the codeassembly store for later retrieval with kb-retrieve-events. ## Quick Start Ask the agent to record the lede decision for the pull request that just merged, then pick a rating when it shows you the agent and merged ledes.
