ship-featurelisted

# Shipping a Feature This skill is the gate between "code written" and "issue closed". It validates the work through targeted lenses (only the ones that apply), captures learnings, and closes the loop with Linear. The flow is **gate-driven**: a classifier inspects the diff and the plan, decides which validation gates apply, and runs only those. A docs-only PR doesn't trigger security review. A schema-only PR doesn't trigger a user-value walkthrough. A search-feature PR triggers the eval suite. This is what keeps validation thorough without crushing velocity. ## Step 0 — Identify the active run If the user didn't specify: find the most recently modified `ai/runs/` folder that has a `plan.md` but no completed `review.md`. Confirm with the user before proceeding. Read: - `ai/runs/<run>/plan.md` — acceptance criteria, user value, efficiency budget, gate scope - `ai/runs/<run>/worklog.md` — deviations and decisions during implementation Set: ```bash RUN_DIR=ai/runs/<run> ``` ## Step 0.5 — Classify gates Run the deterministic classifier. It reads `ai/gates.config.mjs` + the diff + the plan's Gate scope section and writes `gates.manifest.json`. ```bash node tools/gate-classifier.mjs --run "$RUN_DIR" --base origin/main --head HEAD EXIT=$? ``` Print the resulting manifest to the user so they see which gates will run before any reviewer sessions burn tokens. - Exit 0 — manifest written cleanly, proceed. - Exit 2 — discrepancies between plan-declared scope and file-pattern tr