# Sprint > PRD-per-sprint evolution route for an existing harness-built (or brownfield) system — grounds a new PRD against the prior requirement spine, produces a human-approved design amendment against the living specs/design/ baseline, then runs /auto. Companion to /build (sprint 1) and /feature (single-story changes). - Skill: `cwijayasundara/sprint` (Agent Skill) - Install (CLI): `npx skillmds@latest add cwijayasundara/sprint` - Raw SKILL.md: https://api.skillmd.com/api/skills/cwijayasundara/sprint/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Product & Planning - Author: cwijayasundara (https://skillmd.com/u/cwijayasundara) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/cwijayasundara/sprint --- # Sprint Skill — PRD-per-Sprint Evolution `/sprint` is a **thin conductor**, like `/feature`, for evolving a system PRD by PRD without regenerating its architecture from scratch each time. It grounds the new PRD against the prior sprint's requirement spine, amends the living `specs/design/` baseline instead of replacing it, and gates code generation on a human-reviewable diff of the design amendment — never a regenerated document. See `docs/superpowers/specs/2026-07-04-sprint-delta-lane-design.md` for the full design rationale. Use `/build` for sprint 1 (no existing code or specs). Use `/feature` for a single story or small cluster that doesn't need a full PRD. `/sprint` is for "here is the next PRD" — many stories, evolving an existing architecture. **Runs in the main session — do not add `context: fork`.** Like `/feature`, this conductor owns interactive human gates (GATE 1, GATE 2) and delegates the actual work to forked sub-skills. ## Usage ```text /sprint prd-sprint2.md # 2 gates (default) /sprint prd-sprint2.md --autonomous # 1 consolidated gate (folds GATE 1 into GATE 2) ``` There is no `--auto` (zero-gate) mode — GATE 2 (design-delta approval) is never collapsible, by design. ## Wrong-door protection Before anything else, check `specs/design/architecture.md`: - If it does not exist **and** the repo has no source code (a fresh, empty project) — this is sprint 1, not sprint N. Stop and tell the human to run `/build ` instead. - If it does not exist **but** source code exists (a true brownfield app the harness did not build) — proceed to Phase 0's baseline recovery. - If it exists — proceed normally. `/build`'s own Step 0 carries the mirror check: if `specs/design/architecture.md` already exists when `/build` is invoked, `/build` stops and redirects here. ## Phase 0 — Preflight & Baseline (fully automatic, no flags) 1. **Baseline recovery.** If `specs/design/architecture.md` is missing (true brownfield): run `/design --baseline-recovery` (see `design/SKILL.md`'s Baseline Recovery Mode). This runs `/brownfield` discovery first if `specs/brownfield/code-graph.json` does not exist, then derives a living design set from the graph, stamped `provenance: derived-from-code`, with a one-time human approval before proceeding. After this, treat the repo as if it already had an approved baseline. 2. **DeepWiki freshness.** If `specs/brownfield/wiki/` exists, check for a `> STALE since…` banner; if present, patch it incrementally via `/code-map --files` on the flagged files, the same mechanism `/feature` already uses. 3. **Sprint number.** List `specs/brd/sprint-*/` directories; the sprint number for this run is one greater than the highest found, or `2` if only the flat legacy `specs/brd/brd.md` exists (sprint 1 was built before sprint-numbered directories existed). State the resolved sprint number before proceeding. ## Phase 1 — Requirements Delta Run `/brd --delta ` (see `brd/SKILL.md`'s Delta Mode). This writes `specs/brd/sprint-N/` and, critically, `specs/brd/sprint-N/requirements-delta.json` classifying every requirement as new/changed/carried against the prior sprint's spine. Any unresolved `dropped` entry halts here per that skill's Step Δ2 — do not proceed with a silent requirement regression. ## Phase 2 — Story Decomposition Run `/spec specs/brd/sprint-N/brd.md --sprint N` (see `spec/SKILL.md`'s sprint addendum). Writes `specs/stories/sprint-N/`. **Reuse-or-justify — REQUIRED SUB-SKILL: `reuse-or-justify`** when the PRD adds or materially extends behavior. Invoke `reuse-or-justify` at intake with the sprint's theme as the goal string and the sprint's story directory `specs/stories/sprint-N/` as its batch, so intra-batch clusters across the sprint's stories surface, before GATE 1. The sub-skill grounds on reuse-scout itself: on a fire it settles reuse-vs-new (and any touched invariant / budget); otherwise it records the net-new assumption and proceeds. The decision is recorded either way — do not run reuse-scout here yourself. ## GATE 1 — Approve Requirement Delta + Decomposition Present, on one screen: - The requirements-delta classification (new / changed / carried / dropped, with resolution for each dropped item) - The story decomposition summary (epic table, dependency graph, story-point total) from `/spec`'s own Step 7 output Ask: "Does this requirement delta and story decomposition look correct? Approve to proceed to the design amendment, or provide corrections." With `--autonomous`, skip this as a separate stop — fold its summary into the single GATE 2 presentation instead (do not skip the underlying `/brd --delta` and `/spec` grounding gates themselves, only the human stop). ## Phase 3 — Design Delta Run `/design --delta --stories specs/stories/sprint-N/ --amendment-id sprint-N` (see `design/SKILL.md`'s Delta Mode, Steps D1–D6): read the living baseline, spawn one planner agent to write the amendment and amend the living design non-destructively, emit the grounding gate, run the contract-drift check, and run the design-delta evaluator rubric. ## GATE 2 — Approve Design Amendment (never collapsible) This is `design/SKILL.md` Delta Mode's Step D7, run from here. Never skipped, never folded away in any autonomy mode. On approval, the amendment and updated living-design files are committed together in one commit. ## Phase 4 — Tracker Publish (optional) If a tracker is configured (`.claude/tracker-config.json`), run `tracker-publish --granularity group` exactly as `/feature`'s epic lane does. ## Phase 5 — Delta Test Plan Run `/test` (the normal flow, not `--from-cr` — that lane is for a single change-request bug fix with no stories directory) scoped to `specs/stories/sprint-N/`, so every new story gets a proper test plan and grounded verification-matrix entries exactly as a fresh spec/design pass would produce. Any existing area the amendment's Breaking Changes section names is covered by that area's own existing tests, which `/auto`'s normal test gate re-runs — no separate regression-pin pass is needed since sprint-N stories already carry proper `story-traces.json`. ## Phase 6 — Build Run `/auto`. The merged `specs/design/component-map.md` (already updated by Phase 3) means the existing ownership, canvas-sync, layer, and context sensors now enforce the evolved design automatically — no `/auto` changes needed. ## Phase 7 — Gate and PR Run `/gate`, then open the PR(s) exactly as `/build`/`/feature` do. Merge stays human. ## State markers At the start of Phase 0, write `.claude/state/current-sprint` (the resolved sprint number, e.g. `2`) and update `.claude/state/sprint-phase` at the start of every phase (`preflight`, `requirements-delta`, `story-decomposition`, `design-delta`, `tracker-publish`, `test-plan`, `build`, `gate`) so `/status` can show sprint progress: ```bash mkdir -p .claude/state printf '%s' "N" > .claude/state/current-sprint printf '%s' "" > .claude/state/sprint-phase ``` ## Gotchas - **Never let Phase 3 regenerate `specs/design/` from scratch.** If the planner agent's output looks like a fresh design rather than an amendment (missing prior component-map rows, a rewritten architecture.md with no trace to the prior version), stop and re-invoke Delta Mode Step D3 with a stronger instruction to read the baseline first. - **GATE 2 here is not `/feature`'s GATE 2.** They serve different lanes (PRD-scale vs single-story) but share the same underlying `design/SKILL.md` Delta Mode machinery — do not conflate the two conductors. - **Do not skip the requirements-delta dropped-item resolution.** An unresolved `dropped` entry is exactly the silent regression this lane exists to prevent. - **`--autonomous` folds human stops, never machine gates.** The grounding gates, contract-drift check, and design-delta evaluator all still run.