--- name: sprint-close description: 'Make sure to use this skill whenever the user mentions closing a sprint, ending a sprint, sprint retrospective, "we finished the sprint", /sprint-close, wrap up the cycle, or when all committed parcel plans in the active sprint reach COMPLETE. Runs the end-of-sprint ritual: appends the retro into the single sprint.md, triggers a spaghetti-monster scan of everything touched this sprint to refresh REFACTORING.md, captures lessons, moves the sprint.md to .devops/archive/sprints/sprint-{n}-/, and updates SPRINTS.md. This skill CLOSES a sprint — it does not plan one (@sprint-plan) or execute parcels (@pass-the-parcel).' version: 7 updated: 2026-09-18 --- # Sprint Close — Retrospective & Hygiene Ritual > **Boundary:** This skill closes the active sprint. It appends the retro to `sprint.md`, runs the code-quality scan that feeds the next cycle, archives the sprint record, and updates the registers. It does not start planning the next sprint (that's `@sprint-plan`, invoked separately once this is done). ## Prerequisites — Read First 1. `.devops/backlog/SPRINTS.md` — find the ACTIVE sprint row and its sprint link. 2. `.devops/sprints/sprint-{n}-/sprint.md` — the committed scope you're measuring against. 3. Last several entries of `.devops/logs/agent-changelog.md` — what actually shipped this sprint. ## 1. Verify Completion Before Closing Check every plan in the Committed Scope (queue) table: - Each should be COMPLETE (`claim_status: COMPLETE`) and present in `.devops/archive/` (root). - **A plan in `.devops/plans/` with `claim_status: GATE_D_USER_APPROVAL` is NOT complete.** It has been executed to `PHASE_9` but the Gate D verdict has not been recorded and the per-plan wrap-up has not archived it — closing over it would retire a sprint containing an unapproved plan. Treat it as unfinished: either get the Gate D verdict and run its wrap-up first, or record it as **carry-forward** in the retro (Section 4) alongside the plans still in the queue. - If any plan is unfinished, do NOT silently drop it — record it as **carry-forward** in the retro (Section 4). A sprint can close with carry-forward; it just means capacity was over-committed. Unfinished plans still in the sprint queue stay parked for the next planning pass (move them back with `git mv` to `.devops/backlog/--backlog.md` and reset `claim_status: QUEUED`); an executed-but-unverified plan stays in `.devops/plans/` at `GATE_D_USER_APPROVAL` until its verdict lands. Confirm the sprint-level Definition of Done from `sprint.md`: full suite green, lint clean, build exit 0. Run them if not already verified this session. ## 2. Run the Spaghetti Scan (the hygiene step) This is the process trigger that keeps refactoring out of the feature backlog: 1. Determine the set of files touched this sprint (from changelog entries + `git diff --name-only {baseline}`). 2. Invoke `@spaghetti-monster` scoped to those touched areas (or run `scripts/spaghetti-monster-scan.cjs` directly for a quick metric pass). 3. For any file now crossing a threshold (CCN>15 warn / >25 critical, >400 lines, etc. per `.devops/backlog/REFACTORING.md` § Thresholds), add it to the REFACTORING.md Kill List with a note: *"Flagged at close of sprint {n} by {file change}."* 4. Update REFACTORING.md's "Completed Refactors" table for anything this sprint cleaned up. **Scan scope — app source *and* machinery.** `scripts/spaghetti-monster-scan.cjs` walks `src/` (the app surface) **plus** the machinery roots `scripts/`, `.devops/skills/`, `.devops/agents/`, `.devops/templates/`, under the same thresholds. A missing root prints a one-line skip and exits `0` — never a crash, and never a silent skip. Rows whose `imp`/`fn`/`CCN` columns show `-` are non-ECMAScript files (`.ps1`/`.py`/`.md`): they are ranked on **line count only**, which is the load-bearing signal for the machinery pass. Read `-` as "not measured", never as "clean". **The refactoring lane is opt-in per repo.** Step 3's promotion — and § 6's "Ensure REFACTORING.md reflects the scan results" — need `.devops/backlog/REFACTORING.md` to exist (adopted from `.devops/templates/REFACTORING.template.md`). If it does **not** exist, do not create it implicitly: run the scan anyway and write its findings into the retro's **New Refactoring Items** section, stating plainly that the register is absent, so the lane is dormant and the flags have nowhere to land. A skipped promotion is a *recorded* decision — never a silent one. The scan output does NOT create new feature-backlog items. It only feeds REFACTORING.md. ## 3. Measure Capacity Accuracy Compare committed points vs delivered points from `sprint.md`. Record in the retro. This number calibrates future `@sprint-plan` capacity suggestions — it's the single most useful line for making estimates honest over time. ## 4. Append the Retro to sprint.md There is **no separate `retro.md`**. Append a `## Retro` section to the existing `.devops/sprints/sprint-{n}-/sprint.md`, and flip its front-matter `status` to `closed` and `closed:` to today. Fill every section — an empty section is a skipped lesson. ### Appended section template ```markdown ## Retro ### Goal — Met? {Restate the goal from the top of this file. Did we achieve it? Yes / Partially / No, with one-line why.} ### What Shipped | Code | Plan | Size | Effort accuracy | Notes | |------|------|------|-----------------|-------| | {T..} | {title} | {S/M/L} | {estimated vs actual} | {one line} | ### Carry-Forward | Code | Why not done | New size | Next sprint? | |------|--------------|----------|--------------| | {T..} | {reason} | {size} | {Y/N} | ### Metrics: Before → After - Hot spots (>CCN 15): {X} → {Y} - Files >400 lines: {X} → {Y} - Test count: {X} → {Y} - Lint warnings: {X} → {Y} - Capacity: committed {pts} / delivered {pts} = {accuracy %} ### Retro: Keep / Drop / Try - **Keep** (worked, do again): {…} - **Drop** (hurt, stop): {…} - **Try** (next sprint experiment): {…} ### Lessons for the Wiki / Knowledge Capture {Any durable insight worth promoting via @knowledge-capture — app-domain to `.wiki/core/18-knowledge-capture.md`, machinery/process/tooling to `.devops/rules/process-lessons.md`. Reference KC numbers if recorded.} ### New Refactoring Items (→ REFACTORING.md) {List files flagged by the close-of-sprint scan. Confirm they were added to the Kill List.} ``` ## 5. Archive the Sprint Record A closed sprint is a historical record: move the whole sprint folder to the archive. 1. `git mv .devops/sprints/sprint-{n}-/sprint.md .devops/archive/sprints/sprint-{n}-/sprint.md` (create the target folder). 2. Shipped plans are **not** moved here — they already archived to `.devops/archive/` root individually at wrap-up (see `.devops/rules/plan-lifecycle.md` rule 6). They are linked to the sprint by the `sprint:` front-matter field. 3. Remove the now-empty `.devops/sprints/sprint-{n}-/` directory. ## 6. Update the Registers - In `SPRINTS.md`: change the sprint row status to `✅ CLOSED` and set the retro link to the archived `sprint.md` (`.devops/archive/sprints/sprint-{n}-/sprint.md#retro`). Update `last_sprint`. - Ensure REFACTORING.md reflects the scan results (Step 2) — when the register exists; otherwise the retro's **New Refactoring Items** section carries them, with the register's absence stated. - Review `.devops/rules/process-lessons.md`: fold each matured machinery rule into its owning skill or `plan-lifecycle.md` and delete it from the register, so the staging register stays small (~25 entries) instead of becoming a second KC. **Template-side step:** both homes (`.devops/rules/**`, `.devops/skills/**`) are on the portable surface, so a pull replaces them wholesale — a satellite **reports** fold candidates upward and does not attempt the edit locally. - Do NOT auto-open the next sprint. Tell the user the current one is closed and they can run `@sprint-plan` when ready, carrying forward the "capacity accuracy" line so the next plan suggests a better budget. ## 7. Hand Off Summarise: goal met?, what carried forward, top refactoring flags, and the calibrated capacity number for next time. Point the user to `@sprint-plan` to open the next cycle.