[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7E82XFWR30NXZDNJCD8EZ9R":3,"skill-same:01M7E82XFWR30NXZDNJCD8EZ9R":90},{"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":36,"commitSha":37,"skillDirectory":38,"forksCount":14,"repositoryHtmlUrl":39,"repositoryHomepageUrl":6,"ownerHtmlUrl":40,"ownerBlogUrl":41,"ownerDescription":42,"updatedAt":43,"install":44,"resources":57},"01M7E82XFWR30NXZDNJCD8EZ9R","spike-bakeoff",null,"Use when building two to four competing strategies in isolated git worktrees to compare them without landing commits.","tony\u002Fskills","tony","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F26336?v=4","01M72E5RNJ7SQV8F9KBM5RKRMN","Tony Narlock","01M7E7MBN5C6YZ4JJJ5Y82KBJG",0,"MIT","https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fspike-bakeoff\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: spike-bakeoff\ndescription: >-\n  Use when building two to four competing strategies in isolated git\n  worktrees to compare them without landing commits.\nallowed-tools: [\"Bash\", \"Read\", \"Grep\", \"Glob\", \"Edit\", \"Write\", \"AskUserQuestion\", \"Agent\"]\nmetadata:\n  argument-hint: \"[\u003Cgoal>] [--strategies=\\\"a; b; c\\\"] [--prongs=\u003C2-4>] [--keep-trees] [--replay]\"\n  source: \"plugins\u002Fspike\u002Fskills\u002Fbakeoff\u002FSKILL.md\"\n---\n\n# this skill\n\nMulti-strategy spike harness. Where the `spike-probe` skill sends one\ninstrument down one path, a bakeoff enters 2–4 **deliberately\ndifferent strategies** for the same goal, builds each in its own git\nworktree, judges them adversarially, and hands back a verdict plus a\ncommit-by-commit plan for the winner.\n\nEvery contender mutates its worktree freely. No contender touches\nhistory: the worktrees are torn down after judging, and only stashes\n(recoverable by SHA) and the verdict survive.\n\n## Core thesis\n\nWhen the approach is uncertain, building each strategy small and\ncomparing real evidence (diffs, gate results, blast radius, idiom fit)\nbeats arguing in the abstract. A bakeoff is N probes plus a judgment:\neach contender follows the probe discipline (cheapest verification,\n`SPIKE:` markers, stay in goal), and the bakeoff adds isolation,\nblindness (contenders do not see each other), and adversarial judging.\n\nA second bakeoff because the first winner hit new resistance means\nthe `spike-ratchet` skill, which carries a ledger across rounds.\n\n## The Iron Rule\n\n```\nA BAKEOFF PRODUCES ZERO COMMITS\n```\n\nThe zero-commit rule and its rationalization table are in\n`references\u002Fspike-core.md` (Zero commits): no\ncommits in any worktree or on any contender branch, not even to snapshot\na contender. A worktree commit is no more harmless than a trunk commit,\nand `git diff` across stash SHAs compares contenders fine. Losers are\ngraft material: stash every contender with a SHA before any teardown.\n\n**Red flags (STOP):** `git commit` anywhere; `git worktree remove\n--force` on a tree that has not been stashed; \"snapshot commit\";\nmerging a contender branch. All of these mean: stash with a SHA,\nprune, and write the verdict.\n\n## `$ARGUMENTS` contract\n\nNon-flag text is the goal. Resolve it by the ladder below:\n\n1. **Typed goal wins**: non-flag `$ARGUMENTS` text, verbatim.\n2. **Empty → mine the conversation** (review findings, a failing\n   test, an agreed problem). One strong candidate: adopt it (the\n   Phase 1 brief is the confirmation gate). Several:\n   `ask-user-choice`; non-interactive: halt and name the candidates.\n   None: ask; non-interactive: halt and name the missing goal.\n3. **Record provenance** in the Phase 1 brief.\n\n**Strategies** resolve by their own ladder:\n\n1. `--strategies=\"a; b; c\"`: semicolon-separated, verbatim, one\n   contender each.\n2. Otherwise, mine the conversation for competing options (\"we could\n   do A or B\"); they become the contenders, with provenance.\n3. Otherwise, propose 2–4 genuinely distinct strategies in the brief\n   (different architecture, layer, or dependency posture, not cosmetic\n   variants).\n\n| Flag | Default | Effect |\n|---|---|---|\n| `--strategies=\"a; b; c\"` | off | Explicit contender list; skips strategy mining. |\n| `--prongs=\u003C2-4>` | auto | Cap the number of contenders. |\n| `--keep-trees` | off | Skip teardown; leave all worktrees in place for manual inspection. Each contender's work is stashed without being cleared (`stash --keep`), so the tree still shows it, and SHAs are recorded. |\n| `--replay` | off | After the verdict and plan are approved, land the winner immediately: apply its stash, one gated commit per plan item. |\n\n## Phase 0: Situational awareness\n\nThe shared steps live in `references\u002Fspike-core.md`; read it before\nthis phase. In every command, `\u003Cscripts>` stands for the absolute path of\n`..\u002F..\u002Fscripts` resolved from this skill's directory (the shell runs in the\nproject, not here); write that path literally.\n\nRun its Preflight, resolving the gate buckets and CI split per\n`references\u002Fverification-gates.md`. A bakeoff never probes on top of\nthe user's work, because `git worktree add ... HEAD` carries none of it:\noffer only stash first or abort (non-interactive default: abort). It takes\nno `--branch`. Once the brief is approved, run Setup's `save-user` before\nthe first worktree; its printed user-work SHA goes in the report with its\nrestore command. A rejected brief changes nothing. Then:\n\n1. Confirm `git worktree` is usable and there is disk headroom for N\n   checkouts.\n2. Create the worktree root as described under \"Worktree root\" in the\n   shared file.\n3. Note any setup a fresh checkout needs to run gates (dependency\n   install, codegen), because each kitchen must be able to bake.\n\n## Phase 1: Orchestration plan\n\nEnter plan mode before the first edit. In Claude Code call `EnterPlanMode`;\nin Cursor, Codex, or Gemini use `\u002Fplan` or `Shift+Tab`. Where plan mode is\nunavailable, present the plan as text.\n\nPresent the **bakeoff brief**:\n\n1. The goal, one line, with provenance (typed \u002F inferred from what).\n2. The contenders: one line per strategy stating what makes it\n   *distinct*; merge entries that differ only cosmetically. Each is a\n   full probe plus a judge per lens, so state the sub-agent count (N\n   contenders plus the judges) and let the user confirm or trim it.\n3. What \"proven\" means per contender: the shared smoke check every\n   contender must pass to reach judging.\n4. The judging rubric (the lenses in `bakeoff-round.md`, plus any goal-specific\n   criteria the user cares about).\n5. Discovered gate commands and the local-vs-CI split.\n6. Worktree names and root; the exit path: stash-and-teardown \u002F\n   `--keep-trees` \u002F `--replay`.\n\nWait for approval, then exit plan mode. If plan mode is unavailable,\npresent the brief inline and proceed on confirmation. In a\nnon-interactive run, record the brief in the report and proceed.\n\n## Phase 2: The round\n\nRun `references\u002Fbakeoff-round.md` in full: bake, gate, judge, then\nstash and tear down. Its judging lenses and any goal-specific ones come\nfrom the brief's rubric. Then:\n\n1. After a stash-first Phase 0, restore the user's work with\n   `\u003Cscripts>\u002Fspike.sh restore \u003Cuser-work sha>` unless `--replay` runs,\n   which restores it after its last item. A refused restore keeps the\n   stash: report the SHA.\n2. Under `--keep-trees`, name each kept worktree path in the report: the\n   user must act on it (`git worktree remove`).\n\n## Phase 3: The verdict and replay plan\n\nProduce the verdict, then a commit-by-commit plan for the **winner's\nstash**, built per `references\u002Freplay.md` (The replay plan:\nsubjects in the project's commit format, contents per commit,\n`SPIKE:` decisions to resolve, per-commit gates), plus:\n\n- **Grafts**: plan items that pull specific hunks from runner-up\n  stashes (by SHA and file), stated as recommendations. Mark them\n  **unproven in combination**: no contender was built with them. When\n  they are too substantial to land blind, recommend re-probing: apply\n  the winner's stash and the graft hunks, then run\n  the `spike-probe` skill on that tree, answering its dirty-tree halt with\n  *probe on top of it*. That proves winner-plus-grafts as one tree at\n  zero commits.\n- The losing strategies, one line each: why they lost, and under what\n  conditions they would have won.\n\nClose with the local-vs-CI table and the post-push watch command\nwhen observable, as in the replay plan.\n\n## Phase 4: Replay (only with `--replay`, after approval)\n\nFollow `references\u002Freplay.md` (Replay) from the main checkout: apply\nthe winner's stash with `\u003Cscripts>\u002Fspike.sh replay-apply \u003Csha>` (apply,\nnot pop), land plan items one gated commit at a time, apply graft hunks\nfrom runner-up stashes where the plan says so, and only after the final\ngreen gate drop the contender stashes by SHA with\n`\u003Cscripts>\u002Fspike.sh drop`, never the user-work stash. Commit\nmessages follow that reference's rules.\n\nRe-run the shared proving check before committing **each** plan item\nthat carries a graft, not once after the last: a graft in an early\ncommit is already history when a later check fails, and this phase never\nrewrites commits. A graft that fails is dropped from the plan and\nrecorded as dropped; do not debug it here.\n\n## Output contract\n\nWrite the report, and any replay commit message, per\n`references\u002Fprose-rules.md`.\n\nThe first line of the reply is exactly the hero line: no lead-in sentence,\nno progress note sent while sub-agents ran, no first-person opener, no\n\"I recommend\". Print every SHA abbreviated to 7 characters. After the last section\nonly the panel or the default line follows: no table, no closing question\nin prose. Report results, not process: never narrate a\ndeparture from these steps or name the brief or the skill; do the step or\nhalt. State what happened, never what a contender or step did not do: no\n\"not hidden\", \"no user work was saved\", \"nothing was X\" restatements.\nWrite in the passive or the plain result, never \"I\" or \"the run\". Do not\ndescribe how the run was set up (answers unavailable, permissions,\nsandbox); a non-interactive default is one line: \"Default: keep the\nstashes.\" Do not print HEAD or the main checkout's SHA; SHAs belong to\nstashes. Use no transient wording (\"for now\", \"currently\"). Write only the\nsections below: a stray finding goes in the findings or open section, and\nprocess (an amended commit, a retry) is left out.\nReport worktrees as removed only when `git worktree list` no longer shows them.\nWarnings about the checkout (mode changes, files left in place, a tree\nthat needs attention) go inside `## Stashes`, never before the first section.\n\n1. Hero block (1–3 lines): `✓ bakeoff judged: \u003Cwinner>` \u002F\n   `⚠ bakeoff inconclusive` + goal + exit path taken.\n\n   A run that halts before any worktree exists (no commit, repository\n   rooted at home, dirty tree declined, missing code or goal) is a\n   `⚠ bakeoff halted` hero line, one result sentence naming every\n   blocker per `references\u002Fspike-core.md` (Halted run).\n2. `## Contenders`: one line per strategy: what it tried, proven or\n   blocked, headline gate status.\n3. `## Verdict`: one plain line per lens naming each contender's standing\n   (`Correctness: A holds, B ...`, no bold label), then the winner, the\n   graft list, and why the losers lost.\n4. `## Verification`: gate commands run per contender; local-vs-CI\n   split; deferrals.\n5. `## Stashes`: one bullet per contender: name, strategy, stash message,\n   SHA, restore command. Every contender appears, losers included; one\n   that changed nothing shows \"no changes\" in place of the SHA. A run that\n   saved the user's work adds a bullet for it and says whether it was\n   restored.\n6. `## Replay plan`: the numbered commit sequence for the winner,\n   grafts marked.\n7. End with an `ask-user-choice` panel: replay the winner \u002F probe the\n   grafted winner (still zero commits) \u002F keep stashes\n   and stop \u002F discard all, unless already in plan mode or `--replay`\n   was given. Recommend the probe exit when the verdict judged the grafts\n   too substantial to land blind. In a non-interactive run, state the\n   default taken (\"Default: keep the stashes.\") instead of asking.\n\nA halted run replies in this shape:\n\n```text\n⚠ bakeoff halted: compare an LRU dict with a Redis-backed session cache.\n\nBlocked: the working tree has uncommitted changes that a worktree would not carry.\nCommit or stash them yourself, then run spike-bakeoff again.\n```\n\nReply in this shape, as plain text outside any code fence, with plain\nbullets and no bold (never `- **Winner:** ...`), and no local paths:\n\n```text\n✓ bakeoff judged: B (write-behind cache)\nGoal: cut p99 latency of the export endpoint. Exit path: stash and teardown.\n\n## Contenders\n- A, in-process LRU: proven; format, lint, and typecheck pass, 2 tests fail.\n- B, write-behind cache: proven; all gates pass.\n- C, query rewrite: blocked; the planner ignores the new index.\n\n## Verdict\n- Correctness: A and B pass the proving check; C is blocked.\n- Blast radius: B touches two files, A touches five.\n- Idiom fit: B matches the existing cache module.\n- Gate status: A fails two tests; B passes all gates.\n- Winner B. Graft from A: its eviction test.\n\n## Verification\n- Local: format, lint, typecheck, scoped tests. CI-only: integration suite.\n\n## Stashes\n- A, LRU: bakeoff 1-lru: in-process LRU (lost), 9f2c4d1, restore with git stash apply 9f2c4d1\n- B, write-behind: bakeoff 2-write-behind: write-behind cache (won), 5ab3e08, restore with git stash apply 5ab3e08\n- C, query rewrite: bakeoff 3-query: query rewrite (blocked), no changes\n\n## Replay plan\n1. cache(feat[store]) Add write-behind store\n2. cache(test[store]) Cover flush and eviction (graft from A, unproven)\n```\n\n\n## Portability notes\n\n- `ask-user-choice` — present the listed options and wait for the user to pick one. Hosts with a structured multiple-choice tool (Claude Code's `AskUserQuestion`) should use it; otherwise print a numbered list and wait for a numbered reply. Never proceed on an assumed answer.\n- `$ARGUMENTS` — the text the user passed when invoking this skill. If your host does not substitute it, read it as the user's request in the current turn, and ask when there is none.\n- Bundled files — every relative path in this skill points at a file shipped inside this skill directory. Read them from here, not from the host's plugin tree.\n","# this skill\n\nMulti-strategy spike harness. Where the `spike-probe` skill sends one\ninstrument down one path, a bakeoff enters 2–4 **deliberately\ndifferent strategies** for the same goal, builds each in its own git\nworktree, judges them adversarially, and hands back a verdict plus a\ncommit-by-commit plan for the winner.\n\nEvery contender mutates its worktree freely. No contender touches\nhistory: the worktrees are torn down after judging, and only stashes\n(recoverable by SHA) and the verdict survive.\n\n## Core thesis\n\nWhen the approach is uncertain, building each strategy small and\ncomparing real evidence (diffs, gate results, blast radius, idiom fit)\nbeats arguing in the abstract. A bakeoff is N probes plus a judgment:\neach contender follows the probe discipline (cheapest verification,\n`SPIKE:` markers, stay in goal), and the bakeoff adds isolation,\nblindness (contenders do not see each other), and adversarial judging.\n\nA second bakeoff because the first winner hit new resistance means\nthe `spike-ratchet` skill, which carries a ledger across rounds.\n\n## The Iron Rule\n\n```\nA BAKEOFF PRODUCES ZERO COMMITS\n```\n\nThe zero-commit rule and its rationalization table are in\n`references\u002Fspike-core.md` (Zero commits): no\ncommits in any worktree or on any contender branch, not even to snapshot\na contender. A worktree commit is no more harmless than a trunk commit,\nand `git diff` across stash SHAs compares contenders fine. Losers are\ngraft material: stash every contender with a SHA before any teardown.\n\n**Red flags (STOP):** `git commit` anywhere; `git worktree remove\n--force` on a tree that has not been stashed; \"snapshot commit\";\nmerging a contender branch. All of these mean: stash with a SHA,\nprune, and write the verdict.\n\n## `$ARGUMENTS` contract\n\nNon-flag text is the goal. Resolve it by the ladder below:\n\n1. **Typed goal wins**: non-flag `$ARGUMENTS` text, verbatim.\n2. **Empty → mine the conversation** (review findings, a failing\n   test, an agreed problem). One strong candidate: adopt it (the\n   Phase 1 brief is the confirmation gate). Several:\n   `ask-user-choice`; non-interactive: halt and name the candidates.\n   None: ask; non-interactive: halt and name the missing goal.\n3. **Record provenance** in the Phase 1 brief.\n\n**Strategies** resolve by their own ladder:\n\n1. `--strategies=\"a; b; c\"`: semicolon-separated, verbatim, one\n   contender each.\n2. Otherwise, mine the conversation for competing options (\"we could\n   do A or B\"); they become the contenders, with provenance.\n3. Otherwise, propose 2–4 genuinely distinct strategies in the brief\n   (different architecture, layer, or dependency posture, not cosmetic\n   variants).\n\n| Flag | Default | Effect |\n|---|---|---|\n| `--strategies=\"a; b; c\"` | off | Explicit contender list; skips strategy mining. |\n| `--prongs=\u003C2-4>` | auto | Cap the number of contenders. |\n| `--keep-trees` | off | Skip teardown; leave all worktrees in place for manual inspection. Each contender's work is stashed without being cleared (`stash --keep`), so the tree still shows it, and SHAs are recorded. |\n| `--replay` | off | After the verdict and plan are approved, land the winner immediately: apply its stash, one gated commit per plan item. |\n\n## Phase 0: Situational awareness\n\nThe shared steps live in `references\u002Fspike-core.md`; read it before\nthis phase. In every command, `\u003Cscripts>` stands for the absolute path of\n`..\u002F..\u002Fscripts` resolved from this skill's directory (the shell runs in the\nproject, not here); write that path literally.\n\nRun its Preflight, resolving the gate buckets and CI split per\n`references\u002Fverification-gates.md`. A bakeoff never probes on top of\nthe user's work, because `git worktree add ... HEAD` carries none of it:\noffer only stash first or abort (non-interactive default: abort). It takes\nno `--branch`. Once the brief is approved, run Setup's `save-user` before\nthe first worktree; its printed user-work SHA goes in the report with its\nrestore command. A rejected brief changes nothing. Then:\n\n1. Confirm `git worktree` is usable and there is disk headroom for N\n   checkouts.\n2. Create the worktree root as described under \"Worktree root\" in the\n   shared file.\n3. Note any setup a fresh checkout needs to run gates (dependency\n   install, codegen), because each kitchen must be able to bake.\n\n## Phase 1: Orchestration plan\n\nEnter plan mode before the first edit. In Claude Code call `EnterPlanMode`;\nin Cursor, Codex, or Gemini use `\u002Fplan` or `Shift+Tab`. Where plan mode is\nunavailable, present the plan as text.\n\nPresent the **bakeoff brief**:\n\n1. The goal, one line, with provenance (typed \u002F inferred from what).\n2. The contenders: one line per strategy stating what makes it\n   *distinct*; merge entries that differ only cosmetically. Each is a\n   full probe plus a judge per lens, so state the sub-agent count (N\n   contenders plus the judges) and let the user confirm or trim it.\n3. What \"proven\" means per contender: the shared smoke check every\n   contender must pass to reach judging.\n4. The judging rubric (the lenses in `bakeoff-round.md`, plus any goal-specific\n   criteria the user cares about).\n5. Discovered gate commands and the local-vs-CI split.\n6. Worktree names and root; the exit path: stash-and-teardown \u002F\n   `--keep-trees` \u002F `--replay`.\n\nWait for approval, then exit plan mode. If plan mode is unavailable,\npresent the brief inline and proceed on confirmation. In a\nnon-interactive run, record the brief in the report and proceed.\n\n## Phase 2: The round\n\nRun `references\u002Fbakeoff-round.md` in full: bake, gate, judge, then\nstash and tear down. Its judging lenses and any goal-specific ones come\nfrom the brief's rubric. Then:\n\n1. After a stash-first Phase 0, restore the user's work with\n   `\u003Cscripts>\u002Fspike.sh restore \u003Cuser-work sha>` unless `--replay` runs,\n   which restores it after its last item. A refused restore keeps the\n   stash: report the SHA.\n2. Under `--keep-trees`, name each kept worktree path in the report: the\n   user must act on it (`git worktree remove`).\n\n## Phase 3: The verdict and replay plan\n\nProduce the verdict, then a commit-by-commit plan for the **winner's\nstash**, built per `references\u002Freplay.md` (The replay plan:\nsubjects in the project's commit format, contents per commit,\n`SPIKE:` decisions to resolve, per-commit gates), plus:\n\n- **Grafts**: plan items that pull specific hunks from runner-up\n  stashes (by SHA and file), stated as recommendations. Mark them\n  **unproven in combination**: no contender was built with them. When\n  they are too substantial to land blind, recommend re-probing: apply\n  the winner's stash and the graft hunks, then run\n  the `spike-probe` skill on that tree, answering its dirty-tree halt with\n  *probe on top of it*. That proves winner-plus-grafts as one tree at\n  zero commits.\n- The losing strategies, one line each: why they lost, and under what\n  conditions they would have won.\n\nClose with the local-vs-CI table and the post-push watch command\nwhen observable, as in the replay plan.\n\n## Phase 4: Replay (only with `--replay`, after approval)\n\nFollow `references\u002Freplay.md` (Replay) from the main checkout: apply\nthe winner's stash with `\u003Cscripts>\u002Fspike.sh replay-apply \u003Csha>` (apply,\nnot pop), land plan items one gated commit at a time, apply graft hunks\nfrom runner-up stashes where the plan says so, and only after the final\ngreen gate drop the contender stashes by SHA with\n`\u003Cscripts>\u002Fspike.sh drop`, never the user-work stash. Commit\nmessages follow that reference's rules.\n\nRe-run the shared proving check before committing **each** plan item\nthat carries a graft, not once after the last: a graft in an early\ncommit is already history when a later check fails, and this phase never\nrewrites commits. A graft that fails is dropped from the plan and\nrecorded as dropped; do not debug it here.\n\n## Output contract\n\nWrite the report, and any replay commit message, per\n`references\u002Fprose-rules.md`.\n\nThe first line of the reply is exactly the hero line: no lead-in sentence,\nno progress note sent while sub-agents ran, no first-person opener, no\n\"I recommend\". Print every SHA abbreviated to 7 characters. After the last section\nonly the panel or the default line follows: no table, no closing question\nin prose. Report results, not process: never narrate a\ndeparture from these steps or name the brief or the skill; do the step or\nhalt. State what happened, never what a contender or step did not do: no\n\"not hidden\", \"no user work was saved\", \"nothing was X\" restatements.\nWrite in the passive or the plain result, never \"I\" or \"the run\". Do not\ndescribe how the run was set up (answers unavailable, permissions,\nsandbox); a non-interactive default is one line: \"Default: keep the\nstashes.\" Do not print HEAD or the main checkout's SHA; SHAs belong to\nstashes. Use no transient wording (\"for now\", \"currently\"). Write only the\nsections below: a stray finding goes in the findings or open section, and\nprocess (an amended commit, a retry) is left out.\nReport worktrees as removed only when `git worktree list` no longer shows them.\nWarnings about the checkout (mode changes, files left in place, a tree\nthat needs attention) go inside `## Stashes`, never before the first section.\n\n1. Hero block (1–3 lines): `✓ bakeoff judged: \u003Cwinner>` \u002F\n   `⚠ bakeoff inconclusive` + goal + exit path taken.\n\n   A run that halts before any worktree exists (no commit, repository\n   rooted at home, dirty tree declined, missing code or goal) is a\n   `⚠ bakeoff halted` hero line, one result sentence naming every\n   blocker per `references\u002Fspike-core.md` (Halted run).\n2. `## Contenders`: one line per strategy: what it tried, proven or\n   blocked, headline gate status.\n3. `## Verdict`: one plain line per lens naming each contender's standing\n   (`Correctness: A holds, B ...`, no bold label), then the winner, the\n   graft list, and why the losers lost.\n4. `## Verification`: gate commands run per contender; local-vs-CI\n   split; deferrals.\n5. `## Stashes`: one bullet per contender: name, strategy, stash message,\n   SHA, restore command. Every contender appears, losers included; one\n   that changed nothing shows \"no changes\" in place of the SHA. A run that\n   saved the user's work adds a bullet for it and says whether it was\n   restored.\n6. `## Replay plan`: the numbered commit sequence for the winner,\n   grafts marked.\n7. End with an `ask-user-choice` panel: replay the winner \u002F probe the\n   grafted winner (still zero commits) \u002F keep stashes\n   and stop \u002F discard all, unless already in plan mode or `--replay`\n   was given. Recommend the probe exit when the verdict judged the grafts\n   too substantial to land blind. In a non-interactive run, state the\n   default taken (\"Default: keep the stashes.\") instead of asking.\n\nA halted run replies in this shape:\n\n```text\n⚠ bakeoff halted: compare an LRU dict with a Redis-backed session cache.\n\nBlocked: the working tree has uncommitted changes that a worktree would not carry.\nCommit or stash them yourself, then run spike-bakeoff again.\n```\n\nReply in this shape, as plain text outside any code fence, with plain\nbullets and no bold (never `- **Winner:** ...`), and no local paths:\n\n```text\n✓ bakeoff judged: B (write-behind cache)\nGoal: cut p99 latency of the export endpoint. Exit path: stash and teardown.\n\n## Contenders\n- A, in-process LRU: proven; format, lint, and typecheck pass, 2 tests fail.\n- B, write-behind cache: proven; all gates pass.\n- C, query rewrite: blocked; the planner ignores the new index.\n\n## Verdict\n- Correctness: A and B pass the proving check; C is blocked.\n- Blast radius: B touches two files, A touches five.\n- Idiom fit: B matches the existing cache module.\n- Gate status: A fails two tests; B passes all gates.\n- Winner B. Graft from A: its eviction test.\n\n## Verification\n- Local: format, lint, typecheck, scoped tests. CI-only: integration suite.\n\n## Stashes\n- A, LRU: bakeoff 1-lru: in-process LRU (lost), 9f2c4d1, restore with git stash apply 9f2c4d1\n- B, write-behind: bakeoff 2-write-behind: write-behind cache (won), 5ab3e08, restore with git stash apply 5ab3e08\n- C, query rewrite: bakeoff 3-query: query rewrite (blocked), no changes\n\n## Replay plan\n1. cache(feat[store]) Add write-behind store\n2. cache(test[store]) Cover flush and eviction (graft from A, unproven)\n```\n\n\n## Portability notes\n\n- `ask-user-choice` — present the listed options and wait for the user to pick one. Hosts with a structured multiple-choice tool (Claude Code's `AskUserQuestion`) should use it; otherwise print a numbered list and wait for a numbered reply. Never proceed on an assumed answer.\n- `$ARGUMENTS` — the text the user passed when invoking this skill. If your host does not substitute it, read it as the user's request in the current turn, and ask when there is none.\n- Bundled files — every relative path in this skill points at a file shipped inside this skill directory. Read them from here, not from the host's plugin tree.\n",{"name":5,"metadata":24,"description":7,"allowed-tools":27},{"source":25,"argument-hint":26},"plugins\u002Fspike\u002Fskills\u002Fbakeoff\u002FSKILL.md","[\u003Cgoal>] [--strategies=\"a; b; c\"] [--prongs=\u003C2-4>] [--keep-trees] [--replay]",[28,29,30,31,32,33,34,35],"Bash","Read","Grep","Glob","Edit","Write","AskUserQuestion","Agent","main","5fbeb3b47beedea13813be7b7023f736a8d97a65",".agents\u002Fskills\u002Fspike-bakeoff","https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills","https:\u002F\u002Fgithub.com\u002Ftony","https:\u002F\u002Ftony.sh","lighthouse keeper","2026-11-08 17:57:34",{"source":8,"skillName":5,"commands":45,"prompt":55,"zipAvailable":56},[46,49,52],{"tool":47,"command":48},"npx","npx skills add tony\u002Fskills --skill spike-bakeoff",{"tool":50,"command":51},"bunx","bunx skills add tony\u002Fskills --skill spike-bakeoff",{"tool":53,"command":54},"pnpm","pnpm dlx skills add tony\u002Fskills --skill spike-bakeoff","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fspike-bakeoff\u002FSKILL.md",true,[58,64,68,72,76,80,86],{"path":59,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":62,"sourceHtmlUrl":63},"agents\u002Fopenai.yaml","REFERENCE","参考文档",252,"https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fspike-bakeoff\u002Fagents\u002Fopenai.yaml",{"path":65,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":66,"sourceHtmlUrl":67},"references\u002Fbakeoff-round.md",4587,"https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fspike-bakeoff\u002Freferences\u002Fbakeoff-round.md",{"path":69,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":70,"sourceHtmlUrl":71},"references\u002Fprose-rules.md",4644,"https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fspike-bakeoff\u002Freferences\u002Fprose-rules.md",{"path":73,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":74,"sourceHtmlUrl":75},"references\u002Freplay.md",2595,"https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fspike-bakeoff\u002Freferences\u002Freplay.md",{"path":77,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":78,"sourceHtmlUrl":79},"references\u002Fspike-core.md",9755,"https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fspike-bakeoff\u002Freferences\u002Fspike-core.md",{"path":81,"resourceType":82,"resourceTypeDesc":83,"sizeBytes":84,"sourceHtmlUrl":85},"references\u002Fspike.sh","SCRIPT","脚本",11383,"https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fspike-bakeoff\u002Freferences\u002Fspike.sh",{"path":87,"resourceType":60,"resourceTypeDesc":61,"sizeBytes":88,"sourceHtmlUrl":89},"references\u002Fverification-gates.md",4022,"https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fspike-bakeoff\u002Freferences\u002Fverification-gates.md",[91,98,105,112,119,126],{"skillNo":92,"name":93,"nameZh":6,"description":94,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":95,"categories":96,"keywords":97,"publishedAt":19},"01M7E8NCSHXV2JPW4J61R0TQS7","serene-bliss","Use when asked to make a CLI --help, docs page, error message, or tooling UX more serene or calmer for developer experience; returns text, no edits.","https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002Fplugins\u002Fweave\u002Fskills\u002Fserene-bliss\u002FSKILL.md",[],[],{"skillNo":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":102,"categories":103,"keywords":104,"publishedAt":19},"01M7E8N87MAVJD8RRQKQB1ZPS6","review","Weave code review: run independent adversarial reviews in parallel, then synthesize findings","https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002Fplugins\u002Fweave\u002Fskills\u002Freview\u002FSKILL.md",[],[],{"skillNo":106,"name":107,"nameZh":6,"description":108,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":109,"categories":110,"keywords":111,"publishedAt":19},"01M7E8N402XG4ZBFNACQ413163","refine","Use when asked to refine a draft, iterate on code until cleaner, or polish a README through repeated critique and improvement.","https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002Fplugins\u002Fweave\u002Fskills\u002Frefine\u002FSKILL.md",[],[],{"skillNo":113,"name":114,"nameZh":6,"description":115,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":116,"categories":117,"keywords":118,"publishedAt":19},"01M7E8N01KB86A7136P00GVKJW","prompt","Weave prompt: compare independent adversarial implementations, then adopt the single best one","https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002Fplugins\u002Fweave\u002Fskills\u002Fprompt\u002FSKILL.md",[],[],{"skillNo":120,"name":121,"nameZh":6,"description":122,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":123,"categories":124,"keywords":125,"publishedAt":19},"01M7E8MW7YJW2PCMDW3G9RTGNK","plan","Weave planning: compare independent adversarial plans, then synthesize the best one","https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002Fplugins\u002Fweave\u002Fskills\u002Fplan\u002FSKILL.md",[],[],{"skillNo":127,"name":128,"nameZh":6,"description":129,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":130,"categories":131,"keywords":132,"publishedAt":19},"01M7E8MSBYXNCV4ZYWDS4PFJXK","fix-review","Fix weave review findings: validate, add test coverage, fix, and commit each as atomic changes","https:\u002F\u002Fgithub.com\u002Ftony\u002Fskills\u002Fblob\u002Fmain\u002Fplugins\u002Fweave\u002Fskills\u002Ffix-review\u002FSKILL.md",[],[]]