What problem does it solve? When closing a work session or opening a PR, process failures (gates that did not fire, guards with holes, corrections a human had to supply) are easily forgotten or re-described without ever being fixed. This Skill turns those observations into a structured "Session review" section of the PR body where every finding names the mechanism it landed in, or plainly states it landed nowhere. ## Core Features & Use Cases - Ledger-first repeat detection: Reads a beads issue ledger (node scripts/beads-crud.mjs list --label process) before writing, so recurring failures are flagged as repeats with issue IDs and counts instead of looking new each time. - Landing enforcement: Every entry must name a durable mechanism (CI gate, hook glob, drift test, rule file) or explicitly declare itself unlanded; remedies like "be more careful" are rejected. - Gate verification discipline: Requires watching a gate go red before claiming it works, using scripts/mutate.mjs to plant defects and confirm the check catches them. - Use Case: At the end of a debugging session where a human pointed out two missed checks, generate the PR's Session review section with each finding, why nothing caught it, and where the fix permanently landed. ## Quick Start Use the session-review skill to write the Session review section for this PR based on what went wrong in this session.
