[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M75Z1CV08ZXQP4EKRC03YY9F":3,"skill-same:01M75Z1CV08ZXQP4EKRC03YY9F":53},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":14,"repositoryHtmlUrl":27,"repositoryHomepageUrl":28,"ownerHtmlUrl":29,"ownerBlogUrl":30,"ownerDescription":31,"updatedAt":32,"install":33,"resources":46},"01M75Z1CV08ZXQP4EKRC03YY9F","sprint",null,"Run and report on a sprint — fire its waves to completion, check state, close it out. A sprint is a dependency-ordered batch of GitHub issues planned by the `sprint-plan` skill and tagged `sprint-N`. Use when the user says \"fire sprint-N\", \"run sprint-N overnight\", \"what's the state of sprint-N\", \"close sprint-N\", or \"we're on sprint-N\". To PLAN a sprint (decompose + DAG), use `sprint-plan` instead — this skill executes a plan that already exists.","anupam-io\u002Fhive","anupam-io","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F56300182?v=4","01M75Z12MQN0Q1P96EPH6FWM0V","anupam","01M75Z12NC03SNQES4CWTCNXFF",0,"MIT","https:\u002F\u002Fgithub.com\u002Fanupam-io\u002Fhive\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fsprint\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: sprint\ndescription: Run and report on a sprint — fire its waves to completion, check state, close it out. A sprint is a dependency-ordered batch of GitHub issues planned by the `sprint-plan` skill and tagged `sprint-N`. Use when the user says \"fire sprint-N\", \"run sprint-N overnight\", \"what's the state of sprint-N\", \"close sprint-N\", or \"we're on sprint-N\". To PLAN a sprint (decompose + DAG), use `sprint-plan` instead — this skill executes a plan that already exists.\n---\n\n# sprint\n\nThe **runner**. Planning happens in the `sprint-plan` skill, which leaves a\n`promise.md` with a precomputed wave schedule. This skill executes that schedule:\nfire wave after wave until every issue is merged, then close out. The headline use\nis **`run N`** — kick it off and the fleet finishes the sprint unattended (overnight).\n\n- **Sprint** = the labelled batch of issues (`sprint-N` on issue + PR).\n- **Wave** = one parallel fan-out of pods. Waves are precomputed in `promise.md`\n  (the `wave` column), not guessed at fire time.\n\n## Identity & tagging\n\n- `sprint-N` (next integer, never reused) labels every issue and every PR. The\n  driver adds the label to a PR when it ships — workers don't know their sprint.\n- A sprint is scoped **by label**, not branch or milestone. Filter everywhere with\n  `--label sprint-N`.\n- `promise.md` \u002F `result.md` \u002F `dag.html` live in the app repo under\n  `.claude\u002Fsprints\u002Fsprint-N\u002F`, committed to git.\n\n## Sub-actions\n\n### run N — fire to completion (the overnight loop)\n\nThe autonomous driver loop. Requires an approved `promise.md` for sprint N. Runs\nuntil the sprint is done or hits a wall it can't recover. **Safe to re-enter** —\nevery iteration reconciles real state before firing, so a restart never double-fires.\n\n```\nload promise.md → wave schedule + per-issue deps + budget_usd\nloop:\n  0. RECONCILE — if re-entering after a wake\u002Frestart, run `sprint-resume` first so\n     you never fire an issue that already has a live pod or open PR (the Iron Law).\n  1. BUDGET gate (SOFT, not a wall) — sum pod spend + fire count from\n     `hivectl metrics --label sprint-N`. Cap = promise.md `budget_usd` if set, else\n     `HIVE_SPRINT_BUDGET_USD` (200); fire-cap = `HIVE_SPRINT_MAX_AGENTS` (100);\n     hard ceiling = cap × (1 + `HIVE_SPRINT_BUDGET_OVERRUN_PCT`\u002F100). Under the\n     cap: fire freely. In the overrun band (cap→ceiling): keep going but only\n     close-out waves (ship\u002Freview\u002FQA), no big new impl waves. Past the ceiling:\n     STOP firing new work, let in-flight finish, report (AUTO: don't ask). Check\n     before every wave AND every re-fire.\n  2. pick the next unfired wave whose every dep issue is status:done AND PR merged\n  3. fire it — per issue: flip status:ready→in-progress, then `hivectl fire` (see\n     \"fire N\"); append `WAVE k FIRED \u003Cts> issues: …` to .hive\u002Fhandoff\u002Fsprint-N.md\n  4. start the drain watcher; end the turn\n  -- on wake (watcher exit 0, watcher exit 1\u002Ftimeout, or heartbeat) --\n  5. ship APPROVED PRs (Action A); fire reviewers for unreviewed sprint PRs (Action B)\n  6. handle NEEDS_HUMAN per driver.md; on a watcher exit-1, reconcile + reap stuck\n     pods — do NOT assume the wave drained\n  7. not done? → loop (back to 0)\nterminate when: every sprint-N issue is status:done AND every sprint-N PR merged\n  AND QA fold-in is drained or capped (no open `sprint-N` + `type:qa-feedback`\n  issues, OR the `HIVE_QA_FOLD_ROUNDS` \u002F budget ceiling was hit) — see driver.md\n  \"Fold QA findings back into the sprint\"\n  → run `close N`\n```\n\nFire the **whole** wave at once — the planner already capped wave width to cluster\nheadroom (`--max-wave-width`, default 16) and guaranteed file-disjointness, so the\nwave is safe to fan out in one shot. Flipping each issue to `status:in-progress` at\nfire time is what stops a crash-and-resume from re-reading it as `ready` and\ndouble-firing.\n\n**Heartbeat — make it advance, not just report.** The first time `run N` fires a\nwave, start `\u002Floop 5m \u002Fsprint run N` (NOT `status N`). The watcher catches the fast\ndrain; the heartbeat re-enters the *drive* loop on a fixed cadence so a pod that\nhangs without draining still gets shipped \u002F reviewed \u002F reaped instead of merely\nnarrated. Both wakes re-enter `run N`, which reconciles before firing — so they\ncan't double-fire. Start it once per sprint; `close N` stops it.\n\nQA firing is owned by the driver's Action A (it fires the QA bucket when the last\nimpl PR merges) — the run loop does not fire QA independently.\n\n### fire N — one wave\n\nFire a single wave (used by `run N`, or manually to step through):\n\n1. From `promise.md`, pick the lowest-numbered wave not yet fired whose deps are\n   all `status:done`. List its issues.\n2. **Resource sanity check** — how many sprint pods are live\n   (`kubectl -n agents get pods | grep coding-agent | wc -l`)? Will this wave keep\n   the cluster under headroom and the sprint under its cost budget? If tight,\n   shrink or hold.\n3. For each issue: flip it to in-progress, then fire (background, parallel):\n   `gh issue edit \u003CN> --add-label status:in-progress --remove-label status:ready`,\n   then `hivectl fire \u003CN> --type=\u003Chive-type>`. The `promise.md` `type:*` label maps to\n   the `hivectl fire --type=` value:\n\n   | issue label | `hivectl fire --type=` |\n   |---|---|\n   | `type:feature` | `feature-implementation` |\n   | `type:bug` | `bug-fix` |\n   | `type:improvement` | `improvement` |\n   | `type:research` | `research` |\n\n   There are no reviewer waves — each worker reviews and gates its own PR in-pod\n   (status flips to `in-review` on review PASS, `changes-requested` on rework).\n   Re-fire workers on `status:changes-requested` issues (driver Action B).\n   (QA is fired by the driver's Action A via `hivectl qa`, not here.)\n4. **Start the drain watcher**, then end the turn (see below).\n5. Report: issues fired this wave, what's still queued and why, pod count, rough\n   resource posture. End with `RESULT: SPRINT_FIRED N \u003Ccount>`.\n\n### Waiting between waves\n\n`hivectl fire` is fire-and-forget. After firing, sleep on the watcher so the next\nturn wakes when the wave drains:\n\n```bash\nAPP=\"$(hivectl config get app)\" bash \"$HIVE_ROOT\u002F.claude\u002Fskills\u002Fsprint\u002Fwatcher.sh\"   # Bash run_in_background=true\n```\n\nIt selects pods by **label** (`app.kubernetes.io\u002Fname=coding-agent,app=\u003Capp>`) —\nnot by parsing the pod name — and exits **0** when no non-terminal coding-agent pod\nremains. On exit the harness emits a task-notification and the driver wakes. It\nexits **1** on a 90-min timeout or if pods never appeared: treat a non-zero wake as\n\"investigate\" — reconcile via `sprint-resume`, reap\u002Fre-fire stuck pods, and do NOT\nassume the wave drained. Override poll\u002Fcap with `SLEEP=` \u002F `MAX_MIN=`.\n\nThe watcher catches the **fast happy path** (pods finish, wake in seconds); the\n5-minute `run N` heartbeat catches the **slow\u002Fstuck path** (pod hung in Init,\nreviewer looping on cost cap, manual `kubectl` change). Run both — both re-enter\n`run N`, which reconciles before firing, so they can't double-fire.\n\n### status N\n\n| Slice | Query |\n|---|---|\n| Issues | `gh issue list --label sprint-N --state all --json number,title,labels` (bucket by `status:*`) |\n| Open PRs | `gh pr list --label sprint-N --state open --json number,title,labels` (review verdict is the issue's `status:in-review` \u002F `status:changes-requested`) |\n| Merged PRs | `gh pr list --label sprint-N --state merged --json number,title,mergedAt` |\n| Fleet | `hivectl logs` filtered to the sprint's issue numbers |\n\nOutput: elapsed since `promise.md`'s plan timestamp (`2d 4h 12m`; omit if the file\nis missing) · ready \u002F in-review \u002F done counts · open PRs split approved \u002F\nwaiting-review \u002F draft · merged count + which waves remain · **one-line direction\ncall** (\"fire wave 3\" \u002F \"fire reviewers\" \u002F \"fire qa\" \u002F \"ship approved\" \u002F \"sprint\ndone — close\"). End with `RESULT: SPRINT_STATUS N`.\n\n### close N — retro\n\nWrites `result.md`, the close-time artifact that pairs with `promise.md`. The\n**release itself (tag + deploy + verify) is the driver's job** — see the \"Release\"\nsection of `driver.md`. This action produces the retro and triggers that release;\nit does not re-implement it.\n\n1. **Verify done** — all `sprint-N` issues `status:done`, all `sprint-N` PRs\n   merged. If anything is open, bail with the list and suggest `run N` \u002F `status N`.\n   Never close a sprint with open work.\n2. **Load `promise.md`** — the result is judged against it (promised vs delivered,\n   in scope vs drifted). If missing, note that and skip the diff.\n3. **Aggregate metrics** via `hivectl metrics` filtered to the sprint's issue numbers:\n   total cost \u002F tokens \u002F wall time, cost-per-issue avg, slowest + highest-cost\n   issue, breakdown by role (research \u002F impl \u002F reviewer \u002F qa).\n4. **Trigger the release** — hand off to the driver's Release flow (semver bump\n   from what landed → `release\u002F\u003Cversion>` PR → self-merge → tag → `ops\u002Fdeploy.sh`\n   → verify the cluster serves the tagged HEAD). Record the chosen version + deploy\n   outcome for `result.md`.\n5. **Write `result.md`** to `.claude\u002Fsprints\u002Fsprint-N\u002F`:\n   - Header: sprint N, close timestamp (ISO-8601 with time), released version,\n     one-line outcome.\n   - **Promise vs delivered** table: each promised issue ✅ merged \u002F ⚠️ partial \u002F\n     ❌ dropped \u002F ➕ added-mid-sprint.\n   - **Release**: version + one-sentence bump justification + deploy outcome.\n   - **Metrics**: total cost \u002F tokens \u002F agent wall time \u002F **sprint wall time**\n     (close − plan timestamp) \u002F issue count \u002F merged PR count \u002F **wave count**.\n   - **What shipped (Good)**: 3-7 capability highlights (the *capability* a PR\n     added, not its title verbatim).\n   - **What missed (Bad)**: reviewer multi-round pushbacks, NEEDS_INFO bails,\n     cost-cap overruns — what the next sprint should avoid.\n   - **Future plan (Sprint N+1)**: 3-5 bullets — a *proposal*, not a commitment.\n   - **By the numbers**: per-issue cost \u002F wall \u002F tokens \u002F role table.\n6. Stop the sprint's heartbeat loop. **Don't auto-plan N+1** — the future-plan\n   section seeds the user's next `sprint-plan` call.\n7. End with `RESULT: SPRINT_CLOSED N \u003Cversion>`.\n\n> HTML\u002FPDF report: `close` produces `result.md` only. A Range-branded shareable\n> render via `create-html-report` is planned, not yet wired — don't attempt it.\n\n## Coordination with the driver\n\nWhen a sprint is active the driver's per-fire loop still runs, scoped by\n`--label sprint-N`:\n\n| Driver action | Sprint form |\n|---|---|\n| A. Ship reviewed | merge only sprint PRs at `status:in-review` (review PASS); flip issue to `status:done` |\n| B. Resolver loop | re-fire workers on sprint issues at `status:changes-requested` |\n| C. Plan issues | don't plan inside a running sprint — `sprint-plan` owns planning |\n| D. Course-correct | comment on the sprint; don't open a wide new issue mid-sprint |\n\nIf the user hasn't named the active sprint, **ask** — don't default to the\nhighest-numbered label that happens to exist.\n\n## State\n\nActive sprint lives in the session conversation (no persistent active-sprint file).\nPersistent artifacts are the per-sprint files in the app repo under\n`.claude\u002Fsprints\u002Fsprint-N\u002F`: `promise.md` (from `sprint-plan`), `dag.mmd` + `dag.html`\n(the graph), `result.md` (from `close N`).\n\n## Result line\n\nEnd every run with exactly one of:\n- `RESULT: SPRINT_FIRED N \u003Ccount>` — fired a wave (`\u003Ccount>` = pods this wave)\n- `RESULT: SPRINT_STATUS N` — after `status`\n- `RESULT: SPRINT_CLOSED N \u003Cversion>` — after `close`\n- `RESULT: SPRINT_ACTIVE N` — user just set the active sprint; no other action\n- `RESULT: IDLE` — nothing to do this turn\n\nWhen the underlying action was a per-fire driver action (ship\u002Freview\u002Fcorrect),\nemit the driver's result (`SHIPPED` \u002F `PAUSED` \u002F `COURSE_CORRECTED`) instead.\n","# sprint\n\nThe **runner**. Planning happens in the `sprint-plan` skill, which leaves a\n`promise.md` with a precomputed wave schedule. This skill executes that schedule:\nfire wave after wave until every issue is merged, then close out. The headline use\nis **`run N`** — kick it off and the fleet finishes the sprint unattended (overnight).\n\n- **Sprint** = the labelled batch of issues (`sprint-N` on issue + PR).\n- **Wave** = one parallel fan-out of pods. Waves are precomputed in `promise.md`\n  (the `wave` column), not guessed at fire time.\n\n## Identity & tagging\n\n- `sprint-N` (next integer, never reused) labels every issue and every PR. The\n  driver adds the label to a PR when it ships — workers don't know their sprint.\n- A sprint is scoped **by label**, not branch or milestone. Filter everywhere with\n  `--label sprint-N`.\n- `promise.md` \u002F `result.md` \u002F `dag.html` live in the app repo under\n  `.claude\u002Fsprints\u002Fsprint-N\u002F`, committed to git.\n\n## Sub-actions\n\n### run N — fire to completion (the overnight loop)\n\nThe autonomous driver loop. Requires an approved `promise.md` for sprint N. Runs\nuntil the sprint is done or hits a wall it can't recover. **Safe to re-enter** —\nevery iteration reconciles real state before firing, so a restart never double-fires.\n\n```\nload promise.md → wave schedule + per-issue deps + budget_usd\nloop:\n  0. RECONCILE — if re-entering after a wake\u002Frestart, run `sprint-resume` first so\n     you never fire an issue that already has a live pod or open PR (the Iron Law).\n  1. BUDGET gate (SOFT, not a wall) — sum pod spend + fire count from\n     `hivectl metrics --label sprint-N`. Cap = promise.md `budget_usd` if set, else\n     `HIVE_SPRINT_BUDGET_USD` (200); fire-cap = `HIVE_SPRINT_MAX_AGENTS` (100);\n     hard ceiling = cap × (1 + `HIVE_SPRINT_BUDGET_OVERRUN_PCT`\u002F100). Under the\n     cap: fire freely. In the overrun band (cap→ceiling): keep going but only\n     close-out waves (ship\u002Freview\u002FQA), no big new impl waves. Past the ceiling:\n     STOP firing new work, let in-flight finish, report (AUTO: don't ask). Check\n     before every wave AND every re-fire.\n  2. pick the next unfired wave whose every dep issue is status:done AND PR merged\n  3. fire it — per issue: flip status:ready→in-progress, then `hivectl fire` (see\n     \"fire N\"); append `WAVE k FIRED \u003Cts> issues: …` to .hive\u002Fhandoff\u002Fsprint-N.md\n  4. start the drain watcher; end the turn\n  -- on wake (watcher exit 0, watcher exit 1\u002Ftimeout, or heartbeat) --\n  5. ship APPROVED PRs (Action A); fire reviewers for unreviewed sprint PRs (Action B)\n  6. handle NEEDS_HUMAN per driver.md; on a watcher exit-1, reconcile + reap stuck\n     pods — do NOT assume the wave drained\n  7. not done? → loop (back to 0)\nterminate when: every sprint-N issue is status:done AND every sprint-N PR merged\n  AND QA fold-in is drained or capped (no open `sprint-N` + `type:qa-feedback`\n  issues, OR the `HIVE_QA_FOLD_ROUNDS` \u002F budget ceiling was hit) — see driver.md\n  \"Fold QA findings back into the sprint\"\n  → run `close N`\n```\n\nFire the **whole** wave at once — the planner already capped wave width to cluster\nheadroom (`--max-wave-width`, default 16) and guaranteed file-disjointness, so the\nwave is safe to fan out in one shot. Flipping each issue to `status:in-progress` at\nfire time is what stops a crash-and-resume from re-reading it as `ready` and\ndouble-firing.\n\n**Heartbeat — make it advance, not just report.** The first time `run N` fires a\nwave, start `\u002Floop 5m \u002Fsprint run N` (NOT `status N`). The watcher catches the fast\ndrain; the heartbeat re-enters the *drive* loop on a fixed cadence so a pod that\nhangs without draining still gets shipped \u002F reviewed \u002F reaped instead of merely\nnarrated. Both wakes re-enter `run N`, which reconciles before firing — so they\ncan't double-fire. Start it once per sprint; `close N` stops it.\n\nQA firing is owned by the driver's Action A (it fires the QA bucket when the last\nimpl PR merges) — the run loop does not fire QA independently.\n\n### fire N — one wave\n\nFire a single wave (used by `run N`, or manually to step through):\n\n1. From `promise.md`, pick the lowest-numbered wave not yet fired whose deps are\n   all `status:done`. List its issues.\n2. **Resource sanity check** — how many sprint pods are live\n   (`kubectl -n agents get pods | grep coding-agent | wc -l`)? Will this wave keep\n   the cluster under headroom and the sprint under its cost budget? If tight,\n   shrink or hold.\n3. For each issue: flip it to in-progress, then fire (background, parallel):\n   `gh issue edit \u003CN> --add-label status:in-progress --remove-label status:ready`,\n   then `hivectl fire \u003CN> --type=\u003Chive-type>`. The `promise.md` `type:*` label maps to\n   the `hivectl fire --type=` value:\n\n   | issue label | `hivectl fire --type=` |\n   |---|---|\n   | `type:feature` | `feature-implementation` |\n   | `type:bug` | `bug-fix` |\n   | `type:improvement` | `improvement` |\n   | `type:research` | `research` |\n\n   There are no reviewer waves — each worker reviews and gates its own PR in-pod\n   (status flips to `in-review` on review PASS, `changes-requested` on rework).\n   Re-fire workers on `status:changes-requested` issues (driver Action B).\n   (QA is fired by the driver's Action A via `hivectl qa`, not here.)\n4. **Start the drain watcher**, then end the turn (see below).\n5. Report: issues fired this wave, what's still queued and why, pod count, rough\n   resource posture. End with `RESULT: SPRINT_FIRED N \u003Ccount>`.\n\n### Waiting between waves\n\n`hivectl fire` is fire-and-forget. After firing, sleep on the watcher so the next\nturn wakes when the wave drains:\n\n```bash\nAPP=\"$(hivectl config get app)\" bash \"$HIVE_ROOT\u002F.claude\u002Fskills\u002Fsprint\u002Fwatcher.sh\"   # Bash run_in_background=true\n```\n\nIt selects pods by **label** (`app.kubernetes.io\u002Fname=coding-agent,app=\u003Capp>`) —\nnot by parsing the pod name — and exits **0** when no non-terminal coding-agent pod\nremains. On exit the harness emits a task-notification and the driver wakes. It\nexits **1** on a 90-min timeout or if pods never appeared: treat a non-zero wake as\n\"investigate\" — reconcile via `sprint-resume`, reap\u002Fre-fire stuck pods, and do NOT\nassume the wave drained. Override poll\u002Fcap with `SLEEP=` \u002F `MAX_MIN=`.\n\nThe watcher catches the **fast happy path** (pods finish, wake in seconds); the\n5-minute `run N` heartbeat catches the **slow\u002Fstuck path** (pod hung in Init,\nreviewer looping on cost cap, manual `kubectl` change). Run both — both re-enter\n`run N`, which reconciles before firing, so they can't double-fire.\n\n### status N\n\n| Slice | Query |\n|---|---|\n| Issues | `gh issue list --label sprint-N --state all --json number,title,labels` (bucket by `status:*`) |\n| Open PRs | `gh pr list --label sprint-N --state open --json number,title,labels` (review verdict is the issue's `status:in-review` \u002F `status:changes-requested`) |\n| Merged PRs | `gh pr list --label sprint-N --state merged --json number,title,mergedAt` |\n| Fleet | `hivectl logs` filtered to the sprint's issue numbers |\n\nOutput: elapsed since `promise.md`'s plan timestamp (`2d 4h 12m`; omit if the file\nis missing) · ready \u002F in-review \u002F done counts · open PRs split approved \u002F\nwaiting-review \u002F draft · merged count + which waves remain · **one-line direction\ncall** (\"fire wave 3\" \u002F \"fire reviewers\" \u002F \"fire qa\" \u002F \"ship approved\" \u002F \"sprint\ndone — close\"). End with `RESULT: SPRINT_STATUS N`.\n\n### close N — retro\n\nWrites `result.md`, the close-time artifact that pairs with `promise.md`. The\n**release itself (tag + deploy + verify) is the driver's job** — see the \"Release\"\nsection of `driver.md`. This action produces the retro and triggers that release;\nit does not re-implement it.\n\n1. **Verify done** — all `sprint-N` issues `status:done`, all `sprint-N` PRs\n   merged. If anything is open, bail with the list and suggest `run N` \u002F `status N`.\n   Never close a sprint with open work.\n2. **Load `promise.md`** — the result is judged against it (promised vs delivered,\n   in scope vs drifted). If missing, note that and skip the diff.\n3. **Aggregate metrics** via `hivectl metrics` filtered to the sprint's issue numbers:\n   total cost \u002F tokens \u002F wall time, cost-per-issue avg, slowest + highest-cost\n   issue, breakdown by role (research \u002F impl \u002F reviewer \u002F qa).\n4. **Trigger the release** — hand off to the driver's Release flow (semver bump\n   from what landed → `release\u002F\u003Cversion>` PR → self-merge → tag → `ops\u002Fdeploy.sh`\n   → verify the cluster serves the tagged HEAD). Record the chosen version + deploy\n   outcome for `result.md`.\n5. **Write `result.md`** to `.claude\u002Fsprints\u002Fsprint-N\u002F`:\n   - Header: sprint N, close timestamp (ISO-8601 with time), released version,\n     one-line outcome.\n   - **Promise vs delivered** table: each promised issue ✅ merged \u002F ⚠️ partial \u002F\n     ❌ dropped \u002F ➕ added-mid-sprint.\n   - **Release**: version + one-sentence bump justification + deploy outcome.\n   - **Metrics**: total cost \u002F tokens \u002F agent wall time \u002F **sprint wall time**\n     (close − plan timestamp) \u002F issue count \u002F merged PR count \u002F **wave count**.\n   - **What shipped (Good)**: 3-7 capability highlights (the *capability* a PR\n     added, not its title verbatim).\n   - **What missed (Bad)**: reviewer multi-round pushbacks, NEEDS_INFO bails,\n     cost-cap overruns — what the next sprint should avoid.\n   - **Future plan (Sprint N+1)**: 3-5 bullets — a *proposal*, not a commitment.\n   - **By the numbers**: per-issue cost \u002F wall \u002F tokens \u002F role table.\n6. Stop the sprint's heartbeat loop. **Don't auto-plan N+1** — the future-plan\n   section seeds the user's next `sprint-plan` call.\n7. End with `RESULT: SPRINT_CLOSED N \u003Cversion>`.\n\n> HTML\u002FPDF report: `close` produces `result.md` only. A Range-branded shareable\n> render via `create-html-report` is planned, not yet wired — don't attempt it.\n\n## Coordination with the driver\n\nWhen a sprint is active the driver's per-fire loop still runs, scoped by\n`--label sprint-N`:\n\n| Driver action | Sprint form |\n|---|---|\n| A. Ship reviewed | merge only sprint PRs at `status:in-review` (review PASS); flip issue to `status:done` |\n| B. Resolver loop | re-fire workers on sprint issues at `status:changes-requested` |\n| C. Plan issues | don't plan inside a running sprint — `sprint-plan` owns planning |\n| D. Course-correct | comment on the sprint; don't open a wide new issue mid-sprint |\n\nIf the user hasn't named the active sprint, **ask** — don't default to the\nhighest-numbered label that happens to exist.\n\n## State\n\nActive sprint lives in the session conversation (no persistent active-sprint file).\nPersistent artifacts are the per-sprint files in the app repo under\n`.claude\u002Fsprints\u002Fsprint-N\u002F`: `promise.md` (from `sprint-plan`), `dag.mmd` + `dag.html`\n(the graph), `result.md` (from `close N`).\n\n## Result line\n\nEnd every run with exactly one of:\n- `RESULT: SPRINT_FIRED N \u003Ccount>` — fired a wave (`\u003Ccount>` = pods this wave)\n- `RESULT: SPRINT_STATUS N` — after `status`\n- `RESULT: SPRINT_CLOSED N \u003Cversion>` — after `close`\n- `RESULT: SPRINT_ACTIVE N` — user just set the active sprint; no other action\n- `RESULT: IDLE` — nothing to do this turn\n\nWhen the underlying action was a per-fire driver action (ship\u002Freview\u002Fcorrect),\nemit the driver's result (`SHIPPED` \u002F `PAUSED` \u002F `COURSE_CORRECTED`) instead.\n",{"name":5,"description":7},"main","4bcdcf476f67633d1771ef8cd5b4dd2089aa9ee3",".claude\u002Fskills\u002Fsprint","https:\u002F\u002Fgithub.com\u002Fanupam-io\u002Fhive","https:\u002F\u002Fanupam-io.github.io\u002Fhive\u002F","https:\u002F\u002Fgithub.com\u002Fanupam-io","anupam.io","Building @rangesecurity ","2026-06-26 00:09:09",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add anupam-io\u002Fhive --skill sprint",{"tool":39,"command":40},"bunx","bunx skills add anupam-io\u002Fhive --skill sprint",{"tool":42,"command":43},"pnpm","pnpm dlx skills add anupam-io\u002Fhive --skill sprint","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fanupam-io\u002Fhive\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fsprint\u002FSKILL.md",true,[47],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"watcher.sh","SCRIPT","脚本",2250,"https:\u002F\u002Fgithub.com\u002Fanupam-io\u002Fhive\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fsprint\u002Fwatcher.sh",[54,61,68,75,82,89],{"skillNo":55,"name":56,"nameZh":6,"description":57,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":58,"categories":59,"keywords":60,"publishedAt":19},"01M75Z1Z3ED8FY0AVJXWDD2BX4","hive-research","hive's systematic stages for a research fire on a GitHub issue — assess feasibility, map integrations, pin down data formats and APIs, separate what's proven from what's theoretical, and produce a findings document (not code). Follow these stages for any research fire, alongside any app-specific research skill this repo ships.","https:\u002F\u002Fgithub.com\u002Fanupam-io\u002Fhive\u002Fblob\u002Fmain\u002Fk8s-sandbox\u002Fimage\u002Fskills\u002Fhive-research\u002FSKILL.md",[],[],{"skillNo":62,"name":63,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":67,"publishedAt":19},"01M75Z1VXAP9BBK63SJRQR3TZJ","hive-qa","hive's systematic method for a QA fire — verify a running app from the OUTSIDE (you cannot see source code), grounding every claim in observed evidence (baseline capture + live Playwright MCP), then report what's broken, what's working, and what to improve. Follow this for any QA fire.","https:\u002F\u002Fgithub.com\u002Fanupam-io\u002Fhive\u002Fblob\u002Fmain\u002Fk8s-sandbox\u002Fimage\u002Fskills\u002Fhive-qa\u002FSKILL.md",[],[],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":19},"01M75Z1QZ4HPYPAHKZGWGJ1RA3","hive-improvement","Systematic stages for an improvement on a GitHub issue — explore, understand the code, follow conventions, make the targeted improvement without scope creep, test that behaviour is preserved, then leave the tree compiling and lint-clean. Follow these stages in order for any improvement fire.","https:\u002F\u002Fgithub.com\u002Fanupam-io\u002Fhive\u002Fblob\u002Fmain\u002Fk8s-sandbox\u002Fimage\u002Fskills\u002Fhive-improvement\u002FSKILL.md",[],[],{"skillNo":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M75Z1M323T1CK3XZE9NH56RP","hive-feature-implementation","Systematic stages for implementing a feature on a GitHub issue — explore, understand the existing code, follow conventions, implement, test, then leave the tree compiling and lint-clean. Follow these stages in order for any feature-implementation fire.","https:\u002F\u002Fgithub.com\u002Fanupam-io\u002Fhive\u002Fblob\u002Fmain\u002Fk8s-sandbox\u002Fimage\u002Fskills\u002Fhive-feature-implementation\u002FSKILL.md",[],[],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":19},"01M75Z1HAG3X884NB0VS5CRM3D","hive-bug-fix","Systematic stages for fixing a bug on a GitHub issue — explore, understand the code, reproduce and confirm the bug, fix the root cause (not the symptom), test, then leave the tree compiling and lint-clean. Follow these stages in order for any bug-fix fire.","https:\u002F\u002Fgithub.com\u002Fanupam-io\u002Fhive\u002Fblob\u002Fmain\u002Fk8s-sandbox\u002Fimage\u002Fskills\u002Fhive-bug-fix\u002FSKILL.md",[],[],{"skillNo":90,"name":91,"nameZh":6,"description":92,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":93,"categories":94,"keywords":95,"publishedAt":19},"01M75Z18V5CE1JMGRX7XB7XKVV","sprint-resume","Use when re-entering a sprint that may already be running — a new or restarted driver session, a \u002Floop heartbeat waking into a fresh context, after a crash \u002F laptop sleep \u002F context reset, or any time you're unsure which waves are already in flight and firing again could double-fire pods. Reconstructs authoritative wave state from GitHub labels + open PRs + live pods BEFORE any new fire, then hands off to `sprint run N`. Read it before resuming; skip it inside one continuous, uninterrupted run loop.","https:\u002F\u002Fgithub.com\u002Fanupam-io\u002Fhive\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fsprint-resume\u002FSKILL.md",[],[]]