[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6DBC3Z42T7VJFCH2J76MX2Y":3,"skill-same:01M6DBC3Z42T7VJFCH2J76MX2Y":70},{"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":39,"linkOnly":40,"markdownContent":41,"markdownBody":42,"frontMatter":43,"refName":44,"commitSha":45,"skillDirectory":46,"forksCount":14,"repositoryHtmlUrl":47,"repositoryHomepageUrl":6,"ownerHtmlUrl":48,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":49,"install":50,"resources":63},"01M6DBC3Z42T7VJFCH2J76MX2Y","execution-grounded-bon",null,"Use when you have SEVERAL candidate patches\u002Fsolutions for the same task — typically from a cross-vendor Claude+Codex worktree fan-out — and must pick the best one WITHOUT trusting an LLM to judge them (JudgeBench shows model-opinion rerank ≈ chance). Synthesize distinguishing inputs where the candidates diverge, execute all candidates on them, cluster by behavior (behavioral-diff), knock out crashers\u002Ftimeouts, and select by EXECUTION-CONSENSUS — where an independent Claude candidate and an independent Codex candidate agree, that agreement is the signal (cross-vendor = independent failures). A BoN-MAV aspect jury ranks within the winning cluster. Distinct from A3 fail-first (validates ONE author's own tests) and diff-blind-test-synth (one patch vs spec-derived tests). Ships SHADOW: observe + log only, never blocks. Evidence: MASTER A4, COUNCIL C3; arXiv 2502.14382 (S*), 2409.08692 (B4), 2503.15223 (PatchDiff), 2207.10397 (CodeT).","Prekzursil\u002Fagent-skills-toolchain","Prekzursil","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F54636077?v=4","01M6DACZ1PCQHD99SCR1YXG28B","Andrei Mihai Visalon","01M6DACZ670WZF535V3WYYSS4Z",0,"MIT","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fexecution-grounded-bon\u002FSKILL.md",[],[19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38],"execution","execution grounded bon","evidence","test","tests","synthesize","task","diverge","observe","log","patch","first","only","never","diff","use","consensus","pick","select","model","2026-11-10 10:12:30",false,"---\nname: execution-grounded-bon\ndescription: \"Use when you have SEVERAL candidate patches\u002Fsolutions for the same task — typically from a cross-vendor Claude+Codex worktree fan-out — and must pick the best one WITHOUT trusting an LLM to judge them (JudgeBench shows model-opinion rerank ≈ chance). Synthesize distinguishing inputs where the candidates diverge, execute all candidates on them, cluster by behavior (behavioral-diff), knock out crashers\u002Ftimeouts, and select by EXECUTION-CONSENSUS — where an independent Claude candidate and an independent Codex candidate agree, that agreement is the signal (cross-vendor = independent failures). A BoN-MAV aspect jury ranks within the winning cluster. Distinct from A3 fail-first (validates ONE author's own tests) and diff-blind-test-synth (one patch vs spec-derived tests). Ships SHADOW: observe + log only, never blocks. Evidence: MASTER A4, COUNCIL C3; arXiv 2502.14382 (S*), 2409.08692 (B4), 2503.15223 (PatchDiff), 2207.10397 (CodeT).\"\n---\n# Execution-Grounded Best-of-N + Behavioral-Diff Reranker\n\nWhen N agents produce N candidate patches for the same task, the naive rerank is to ask a model \"which is best?\" — and that is proven near-worthless: an LLM judge over candidate code lands around chance on adversarial selection (JudgeBench), and it is exactly the same-context self-grading the whole Theme-A verification program exists to route around (Goodhart + the intrinsic self-correction ceiling, Huang ICLR'24). The done-signal must be **external and agent-inauthorable**.\n\nExecution is that signal. Instead of an opinion, run every candidate on inputs chosen to make them disagree, watch what they actually *do*, and let their behavior cluster them. Candidates that crash or hang eliminate themselves. Candidates that independently converge on the same behavior — especially across two different vendors — are the ones to trust: two independent samples agreeing is the classic oracle-free consensus signal (CodeT), and cross-vendor agreement makes the two samples' blind spots independent (a Claude bug and a Codex bug are unlikely to be the *same* bug).\n\nGrounded in: **S\\*** (2502.14382, distinguishing-input synthesis for candidate selection), **B4** (2409.08692, Bayesian best-selection over execution results), **PatchDiff** (2503.15223, behavioral patch differencing), **CodeT** (2207.10397, dual-execution agreement without a ground-truth oracle). MASTER A4 \u002F COUNCIL C3 both mandate: gate on executable evidence, **never model opinion**.\n\n## When to run\n- You fanned out the same task to ≥2 candidates (Claude subagents in worktrees, and\u002For Codex threads via `mcp__codex__codex`) and need to pick ONE to keep.\n- A fleet\u002Forchestrated work-unit produced competing patches and \"they all pass their own tests\" is not enough to choose.\n- Best-of-N sampling of a single model where you want to select the winner by behavior, not by re-prompting a judge.\n- NOT for: a single patch (use `diff-blind-test-synth` for the spec-vs-patch oracle, or `refactor-equivalence` for pure refactors); choosing between designs\u002Fprose with no executable behavior (there is nothing to execute — this skill does not apply).\n\n## What makes this distinct (do not conflate)\n- **vs A3 fail-first \u002F red-proof:** that proves ONE author's own changed tests fail on the merge-base. Here there are MANY candidates and the signal is cross-candidate behavioral agreement, no author-authored test involved.\n- **vs `diff-blind-test-synth`:** that judges ONE patch against tests written blind to the diff. Here we judge N patches against each other by execution — no spec-derived test needed, the candidates *are* each other's oracle.\n- **vs an LLM-judge rerank:** explicitly rejected. No model scores the candidates. Every ranking key (knockout, cluster size, vendor diversity, aspect jury) is computed from process exit + captured output, nothing a candidate's author can talk its way past.\n\n## The bus this rides (do not rebuild)\n- **Worktree isolation** — each candidate already lives in \u002F can be checked out to its own worktree (the existing `WorktreeCreate` \u002F `Agent(isolation:worktree)` flow). The reranker runs each candidate's command in that candidate's `cwd`; it never edits any tree.\n- **Cross-vendor fan-out** — Claude candidates from parallel `Agent()` subagents; Codex candidates from parallel `mcp__codex__codex` threads. Tag each candidate's `vendor` so the consensus stage can reward independence.\n- **Event ledger** — the verdict is logged (shadow), not enforced; `note.py` or `--ledger` records the `BON_RERANK` line.\n\n## Protocol\n1. **Collect the candidates.** For each, record: a stable `id`, its `vendor` (`claude` | `codex` | …), and either its `cwd`+`cmd` (a deterministic command that consumes one probe input and emits observable output) or precomputed `observations`. Optionally `diffLines` (for the parsimony juror).\n2. **Synthesize distinguishing inputs** (the crux — S\\* \u002F PatchDiff). You want inputs where the candidates *diverge*, because inputs they all handle identically carry no selection signal. Practical recipe:\n   - Ask each candidate (or a fresh agent) for 1–2 inputs that would **break the other candidates** — adversarial boundary cases, off-by-one edges, empty\u002Foversized\u002Fmalformed inputs, the exact case the task's bug was about.\n   - Union those with the task's existing tests \u002F acceptance inputs.\n   - Keep them **deterministic** (no clock\u002Frandom\u002Fnetwork in the probe path) so behavior is reproducible.\n   - You do not need a correct answer for any probe — consensus, not an oracle, decides.\n3. **Execute all candidates on all probes.** `rerank.mjs --execute` runs each `cmd` once per probe in its `cwd`, capturing exit code + stdout with a per-candidate timeout. (Or run them yourself and hand in `observations`.)\n4. **Read the verdict** (below). It clusters by behavior, knocks out crashers\u002Ftimeouts, and picks the cross-vendor-consensus winner, with a confidence grade and a `discrimination` score telling you whether your probes actually separated the field.\n5. **Act on confidence, log always.** `high` (cross-vendor consensus + real discrimination) → adopt the winner with a note. `low` \u002F `indistinguishable` → your probes are too weak or the pool is single-vendor: synthesize sharper distinguishing inputs (step 2) or add a candidate from the other vendor. Record `BON_RERANK:\u003Cwinner> confidence=\u003C…>` to the ledger. **Never auto-block** (SHADOW).\n\n## The reranker: `rerank.mjs`\nDependency-free Node ESM (stdlib only, Node ≥18). Two modes; both produce the same verdict shape.\n\n```bash\n# precomputed: manifest already carries candidates[].observations[]\nnode rerank.mjs --manifest candidates.json\n\n# execute: run each candidate's cmd once per probe to build the behavior matrix\nnode rerank.mjs --manifest candidates.json --execute --ledger .beads\u002Fbon-shadow.jsonl\n\n# machine-readable (pure JSON on stdout; terminal marker goes to stderr)\nnode rerank.mjs --manifest candidates.json --json\n\n# prove the logic on this machine (13 embedded assertions)\nnode rerank.mjs --self-test\n```\n\n### Manifest schema\n```jsonc\n{\n  \"task\": \"fix-null-deref\",\n  \"normalize\": [ { \"pattern\": \"ts=\\\\d+\", \"flags\": \"g\", \"replace\": \"ts=\u003Cn>\" } ], \u002F\u002F strip volatile tokens before hashing (optional)\n  \"tiebreak\": [\"vendors\", \"size\", \"jury\"],   \u002F\u002F cluster ranking order (default)\n  \"probes\": [\n    { \"id\": \"p1\", \"input\": \"…\", \"note\": \"distinguishing input where candidates diverge\" }\n  ],\n  \"candidates\": [\n    \u002F\u002F EXECUTE mode: give cmd (+ optional cwd, probeArg, timeoutMs, diffLines)\n    { \"id\": \"claudeA\", \"vendor\": \"claude\", \"cwd\": \"\u002Fwt\u002Fa\", \"cmd\": [\"node\",\"solve.mjs\"], \"probeArg\": \"stdin\", \"timeoutMs\": 20000, \"diffLines\": 12 },\n    \u002F\u002F PRECOMPUTED mode: give observations (one per probe)\n    { \"id\": \"codexB\",  \"vendor\": \"codex\",\n      \"observations\": [ { \"probe\": \"p1\", \"status\": \"ok\", \"exit\": 0, \"stdout\": \"42\", \"stdout2\": \"42\" } ] }\n  ]\n}\n```\n- `probeArg`: `\"stdin\"` (default — probe `input` piped to stdin) or `\"arg\"` (appended as the last argv item).\n- `observations[].status`: `ok` (ran to a graceful exit, any code) | `error` (couldn't spawn) | `timeout` | `crash` (killed by signal). The three non-`ok` statuses trigger knockout.\n- `observations[].stdout2` (optional): a second run's output — enables the `determinism` juror (flags nondeterministic candidates).\n\n### Pipeline (all stages pure & execution-derived — no model opinion)\n1. **observe** — build the candidate × probe matrix (execute the `cmd`s, or read `observations`).\n2. **knockout** — any candidate that crashes \u002F times out \u002F errors on *any* probe is eliminated (strictly-broken; a hard, oracle-free signal). It cannot cluster or win.\n3. **cluster (behavioral-diff)** — survivors are grouped into equivalence classes by their behavior vector `[status, exit, normalized-stdout]` per probe (sha256 key). Same behavior ⇒ same cluster.\n4. **consensus** — clusters are ranked by, in order: **distinct vendors** (cross-vendor agreement first — a 2-vendor cluster beats a larger 1-vendor cluster because independent samples agreeing is the strongest signal), then **size**, then the winning cluster's **best aspect-jury score**.\n5. **BoN-MAV aspect jury** — within the winning cluster, candidates are ranked by execution-derived jurors (each in [0,1], weights renormalized over present aspects): `probeCompletion`, `outputSanity`, `determinism` (if a repeat run was given), `parsimony` (smaller diff, if `diffLines` given), `latency` (if timed). The top-scoring member is the `winner`.\n\n### Verdict (excerpt)\n```json\n{\n  \"ok\": true, \"task\": \"…\", \"winner\": \"claudeA\", \"winnerVendor\": \"claude\",\n  \"winnerCluster\": \"c1\", \"confidence\": \"high\", \"discrimination\": 1.0,\n  \"indistinguishable\": false,\n  \"clusters\": [ { \"id\": \"c1\", \"members\": [\"claudeA\",\"codexA\"], \"vendors\": [\"claude\",\"codex\"], \"size\": 2 } ],\n  \"knockedOut\": [ { \"id\": \"codexB\", \"reason\": \"timeout on probe p5\" } ],\n  \"reasons\": [ \"cross-vendor consensus (independent failures) — strongest signal\" ]\n}\n```\n- **confidence**: `high` = cross-vendor winning cluster + `discrimination ≥ 0.3` + ≥2 survivors; `low` = single survivor, single-vendor pool, or all-indistinguishable; `medium` otherwise; `none` = empty pool. It is a mechanical function of the execution facts, not a model's opinion.\n- **discrimination** = fraction of probes on which survivors diverge. Near 0 ⇒ your probes don't separate the candidates ⇒ go back to step 2 and synthesize sharper distinguishing inputs. This number is the honest limit indicator: the reranker cannot tell apart candidates your probes cannot tell apart.\n- **indistinguishable** = all survivors landed in one cluster ⇒ behaviorally identical on the given probes; adopt any of them but at `low` confidence until sharper probes exist.\n\n## SHADOW mode (per COUNCIL C5 — always ship observe-only)\nDefault behavior **never blocks and never nags**: `rerank.mjs` always exits 0, emits the verdict + a `SUCCESS:rerank-\u003Ctask> …` terminal marker, and (with `--ledger`) appends one JSONL shadow record. A reranker crash or bad manifest also fails open (exit 0, `FAILED:…` note). The `--enforce` flag — SHADOW-**off**, do not ship on by default — turns a `low`\u002F`none`-confidence verdict into exit 2 so a caller *could* gate on it; graduate OFF→SHADOW→ADVISE→ENFORCE only after a burn-in shows the winner-adoption rate and false-knockout rate are trustworthy on your workload.\n\n## Anti-overfit \u002F honest limits\n- **No trained critic, no vector store, no second runtime** — pure stdlib clustering over the existing worktree\u002FCodex fan-out bus. Stays on the anti-overfit rail the council charter requires.\n- **No oracle is claimed.** Consensus can be confidently wrong if all candidates share a bug (the reason single-vendor pools are capped at `low`, and cross-vendor is rewarded). If every candidate is wrong the same way, `discrimination` is low and the verdict says so — it does not invent a right answer.\n- **Distinguishing-input quality is the ceiling.** Weak probes ⇒ low discrimination ⇒ low confidence; the reranker surfaces this rather than papering over it. Investing in step 2 is where the signal comes from.\n- **Determinism required in the probe path.** A candidate whose output depends on the clock\u002Frandom\u002Fnetwork will look nondeterministic; use `normalize` to strip known-volatile tokens, or provide `stdout2` so the determinism juror can down-rank genuinely flaky candidates.\n- **Knockout is for hard failures only** (crash\u002Ftimeout\u002Ferror), never for a \"wrong-looking\" answer — there is no oracle to call an answer wrong, so a non-zero exit that is graceful and identical across candidates is legitimate agreement, not a knockout.\n","# Execution-Grounded Best-of-N + Behavioral-Diff Reranker\n\nWhen N agents produce N candidate patches for the same task, the naive rerank is to ask a model \"which is best?\" — and that is proven near-worthless: an LLM judge over candidate code lands around chance on adversarial selection (JudgeBench), and it is exactly the same-context self-grading the whole Theme-A verification program exists to route around (Goodhart + the intrinsic self-correction ceiling, Huang ICLR'24). The done-signal must be **external and agent-inauthorable**.\n\nExecution is that signal. Instead of an opinion, run every candidate on inputs chosen to make them disagree, watch what they actually *do*, and let their behavior cluster them. Candidates that crash or hang eliminate themselves. Candidates that independently converge on the same behavior — especially across two different vendors — are the ones to trust: two independent samples agreeing is the classic oracle-free consensus signal (CodeT), and cross-vendor agreement makes the two samples' blind spots independent (a Claude bug and a Codex bug are unlikely to be the *same* bug).\n\nGrounded in: **S\\*** (2502.14382, distinguishing-input synthesis for candidate selection), **B4** (2409.08692, Bayesian best-selection over execution results), **PatchDiff** (2503.15223, behavioral patch differencing), **CodeT** (2207.10397, dual-execution agreement without a ground-truth oracle). MASTER A4 \u002F COUNCIL C3 both mandate: gate on executable evidence, **never model opinion**.\n\n## When to run\n- You fanned out the same task to ≥2 candidates (Claude subagents in worktrees, and\u002For Codex threads via `mcp__codex__codex`) and need to pick ONE to keep.\n- A fleet\u002Forchestrated work-unit produced competing patches and \"they all pass their own tests\" is not enough to choose.\n- Best-of-N sampling of a single model where you want to select the winner by behavior, not by re-prompting a judge.\n- NOT for: a single patch (use `diff-blind-test-synth` for the spec-vs-patch oracle, or `refactor-equivalence` for pure refactors); choosing between designs\u002Fprose with no executable behavior (there is nothing to execute — this skill does not apply).\n\n## What makes this distinct (do not conflate)\n- **vs A3 fail-first \u002F red-proof:** that proves ONE author's own changed tests fail on the merge-base. Here there are MANY candidates and the signal is cross-candidate behavioral agreement, no author-authored test involved.\n- **vs `diff-blind-test-synth`:** that judges ONE patch against tests written blind to the diff. Here we judge N patches against each other by execution — no spec-derived test needed, the candidates *are* each other's oracle.\n- **vs an LLM-judge rerank:** explicitly rejected. No model scores the candidates. Every ranking key (knockout, cluster size, vendor diversity, aspect jury) is computed from process exit + captured output, nothing a candidate's author can talk its way past.\n\n## The bus this rides (do not rebuild)\n- **Worktree isolation** — each candidate already lives in \u002F can be checked out to its own worktree (the existing `WorktreeCreate` \u002F `Agent(isolation:worktree)` flow). The reranker runs each candidate's command in that candidate's `cwd`; it never edits any tree.\n- **Cross-vendor fan-out** — Claude candidates from parallel `Agent()` subagents; Codex candidates from parallel `mcp__codex__codex` threads. Tag each candidate's `vendor` so the consensus stage can reward independence.\n- **Event ledger** — the verdict is logged (shadow), not enforced; `note.py` or `--ledger` records the `BON_RERANK` line.\n\n## Protocol\n1. **Collect the candidates.** For each, record: a stable `id`, its `vendor` (`claude` | `codex` | …), and either its `cwd`+`cmd` (a deterministic command that consumes one probe input and emits observable output) or precomputed `observations`. Optionally `diffLines` (for the parsimony juror).\n2. **Synthesize distinguishing inputs** (the crux — S\\* \u002F PatchDiff). You want inputs where the candidates *diverge*, because inputs they all handle identically carry no selection signal. Practical recipe:\n   - Ask each candidate (or a fresh agent) for 1–2 inputs that would **break the other candidates** — adversarial boundary cases, off-by-one edges, empty\u002Foversized\u002Fmalformed inputs, the exact case the task's bug was about.\n   - Union those with the task's existing tests \u002F acceptance inputs.\n   - Keep them **deterministic** (no clock\u002Frandom\u002Fnetwork in the probe path) so behavior is reproducible.\n   - You do not need a correct answer for any probe — consensus, not an oracle, decides.\n3. **Execute all candidates on all probes.** `rerank.mjs --execute` runs each `cmd` once per probe in its `cwd`, capturing exit code + stdout with a per-candidate timeout. (Or run them yourself and hand in `observations`.)\n4. **Read the verdict** (below). It clusters by behavior, knocks out crashers\u002Ftimeouts, and picks the cross-vendor-consensus winner, with a confidence grade and a `discrimination` score telling you whether your probes actually separated the field.\n5. **Act on confidence, log always.** `high` (cross-vendor consensus + real discrimination) → adopt the winner with a note. `low` \u002F `indistinguishable` → your probes are too weak or the pool is single-vendor: synthesize sharper distinguishing inputs (step 2) or add a candidate from the other vendor. Record `BON_RERANK:\u003Cwinner> confidence=\u003C…>` to the ledger. **Never auto-block** (SHADOW).\n\n## The reranker: `rerank.mjs`\nDependency-free Node ESM (stdlib only, Node ≥18). Two modes; both produce the same verdict shape.\n\n```bash\n# precomputed: manifest already carries candidates[].observations[]\nnode rerank.mjs --manifest candidates.json\n\n# execute: run each candidate's cmd once per probe to build the behavior matrix\nnode rerank.mjs --manifest candidates.json --execute --ledger .beads\u002Fbon-shadow.jsonl\n\n# machine-readable (pure JSON on stdout; terminal marker goes to stderr)\nnode rerank.mjs --manifest candidates.json --json\n\n# prove the logic on this machine (13 embedded assertions)\nnode rerank.mjs --self-test\n```\n\n### Manifest schema\n```jsonc\n{\n  \"task\": \"fix-null-deref\",\n  \"normalize\": [ { \"pattern\": \"ts=\\\\d+\", \"flags\": \"g\", \"replace\": \"ts=\u003Cn>\" } ], \u002F\u002F strip volatile tokens before hashing (optional)\n  \"tiebreak\": [\"vendors\", \"size\", \"jury\"],   \u002F\u002F cluster ranking order (default)\n  \"probes\": [\n    { \"id\": \"p1\", \"input\": \"…\", \"note\": \"distinguishing input where candidates diverge\" }\n  ],\n  \"candidates\": [\n    \u002F\u002F EXECUTE mode: give cmd (+ optional cwd, probeArg, timeoutMs, diffLines)\n    { \"id\": \"claudeA\", \"vendor\": \"claude\", \"cwd\": \"\u002Fwt\u002Fa\", \"cmd\": [\"node\",\"solve.mjs\"], \"probeArg\": \"stdin\", \"timeoutMs\": 20000, \"diffLines\": 12 },\n    \u002F\u002F PRECOMPUTED mode: give observations (one per probe)\n    { \"id\": \"codexB\",  \"vendor\": \"codex\",\n      \"observations\": [ { \"probe\": \"p1\", \"status\": \"ok\", \"exit\": 0, \"stdout\": \"42\", \"stdout2\": \"42\" } ] }\n  ]\n}\n```\n- `probeArg`: `\"stdin\"` (default — probe `input` piped to stdin) or `\"arg\"` (appended as the last argv item).\n- `observations[].status`: `ok` (ran to a graceful exit, any code) | `error` (couldn't spawn) | `timeout` | `crash` (killed by signal). The three non-`ok` statuses trigger knockout.\n- `observations[].stdout2` (optional): a second run's output — enables the `determinism` juror (flags nondeterministic candidates).\n\n### Pipeline (all stages pure & execution-derived — no model opinion)\n1. **observe** — build the candidate × probe matrix (execute the `cmd`s, or read `observations`).\n2. **knockout** — any candidate that crashes \u002F times out \u002F errors on *any* probe is eliminated (strictly-broken; a hard, oracle-free signal). It cannot cluster or win.\n3. **cluster (behavioral-diff)** — survivors are grouped into equivalence classes by their behavior vector `[status, exit, normalized-stdout]` per probe (sha256 key). Same behavior ⇒ same cluster.\n4. **consensus** — clusters are ranked by, in order: **distinct vendors** (cross-vendor agreement first — a 2-vendor cluster beats a larger 1-vendor cluster because independent samples agreeing is the strongest signal), then **size**, then the winning cluster's **best aspect-jury score**.\n5. **BoN-MAV aspect jury** — within the winning cluster, candidates are ranked by execution-derived jurors (each in [0,1], weights renormalized over present aspects): `probeCompletion`, `outputSanity`, `determinism` (if a repeat run was given), `parsimony` (smaller diff, if `diffLines` given), `latency` (if timed). The top-scoring member is the `winner`.\n\n### Verdict (excerpt)\n```json\n{\n  \"ok\": true, \"task\": \"…\", \"winner\": \"claudeA\", \"winnerVendor\": \"claude\",\n  \"winnerCluster\": \"c1\", \"confidence\": \"high\", \"discrimination\": 1.0,\n  \"indistinguishable\": false,\n  \"clusters\": [ { \"id\": \"c1\", \"members\": [\"claudeA\",\"codexA\"], \"vendors\": [\"claude\",\"codex\"], \"size\": 2 } ],\n  \"knockedOut\": [ { \"id\": \"codexB\", \"reason\": \"timeout on probe p5\" } ],\n  \"reasons\": [ \"cross-vendor consensus (independent failures) — strongest signal\" ]\n}\n```\n- **confidence**: `high` = cross-vendor winning cluster + `discrimination ≥ 0.3` + ≥2 survivors; `low` = single survivor, single-vendor pool, or all-indistinguishable; `medium` otherwise; `none` = empty pool. It is a mechanical function of the execution facts, not a model's opinion.\n- **discrimination** = fraction of probes on which survivors diverge. Near 0 ⇒ your probes don't separate the candidates ⇒ go back to step 2 and synthesize sharper distinguishing inputs. This number is the honest limit indicator: the reranker cannot tell apart candidates your probes cannot tell apart.\n- **indistinguishable** = all survivors landed in one cluster ⇒ behaviorally identical on the given probes; adopt any of them but at `low` confidence until sharper probes exist.\n\n## SHADOW mode (per COUNCIL C5 — always ship observe-only)\nDefault behavior **never blocks and never nags**: `rerank.mjs` always exits 0, emits the verdict + a `SUCCESS:rerank-\u003Ctask> …` terminal marker, and (with `--ledger`) appends one JSONL shadow record. A reranker crash or bad manifest also fails open (exit 0, `FAILED:…` note). The `--enforce` flag — SHADOW-**off**, do not ship on by default — turns a `low`\u002F`none`-confidence verdict into exit 2 so a caller *could* gate on it; graduate OFF→SHADOW→ADVISE→ENFORCE only after a burn-in shows the winner-adoption rate and false-knockout rate are trustworthy on your workload.\n\n## Anti-overfit \u002F honest limits\n- **No trained critic, no vector store, no second runtime** — pure stdlib clustering over the existing worktree\u002FCodex fan-out bus. Stays on the anti-overfit rail the council charter requires.\n- **No oracle is claimed.** Consensus can be confidently wrong if all candidates share a bug (the reason single-vendor pools are capped at `low`, and cross-vendor is rewarded). If every candidate is wrong the same way, `discrimination` is low and the verdict says so — it does not invent a right answer.\n- **Distinguishing-input quality is the ceiling.** Weak probes ⇒ low discrimination ⇒ low confidence; the reranker surfaces this rather than papering over it. Investing in step 2 is where the signal comes from.\n- **Determinism required in the probe path.** A candidate whose output depends on the clock\u002Frandom\u002Fnetwork will look nondeterministic; use `normalize` to strip known-volatile tokens, or provide `stdout2` so the determinism juror can down-rank genuinely flaky candidates.\n- **Knockout is for hard failures only** (crash\u002Ftimeout\u002Ferror), never for a \"wrong-looking\" answer — there is no oracle to call an answer wrong, so a non-zero exit that is graceful and identical across candidates is legitimate agreement, not a knockout.\n",{"name":5,"description":7},"main","a298c929b6cbd662e1fce54302ce520d5ceea1a5",".agents\u002Fskills\u002Fexecution-grounded-bon","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain","https:\u002F\u002Fgithub.com\u002FPrekzursil","2026-11-02 14:30:57",{"source":8,"skillName":5,"commands":51,"prompt":61,"zipAvailable":62},[52,55,58],{"tool":53,"command":54},"npx","npx skills add Prekzursil\u002Fagent-skills-toolchain --skill execution-grounded-bon",{"tool":56,"command":57},"bunx","bunx skills add Prekzursil\u002Fagent-skills-toolchain --skill execution-grounded-bon",{"tool":59,"command":60},"pnpm","pnpm dlx skills add Prekzursil\u002Fagent-skills-toolchain --skill execution-grounded-bon","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fexecution-grounded-bon\u002FSKILL.md",true,[64],{"path":65,"resourceType":66,"resourceTypeDesc":67,"sizeBytes":68,"sourceHtmlUrl":69},"rerank.mjs","SCRIPT","脚本",25395,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fexecution-grounded-bon\u002Frerank.mjs",[71,90,120,147,161,177],{"skillNo":72,"name":73,"nameZh":6,"description":74,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":75,"categories":76,"keywords":77,"publishedAt":39},"01M6DF7D7478R55P46REYH4QAC","clarify-requirements","Resolve genuine pre-execution ambiguity by scanning repo context, generating 1-6 grounded options, and asking 1-3 focused questions via AskUserQuestion. NOT for mid-task use (injection risk); NOT when the goal is already unambiguous; use to disambiguate vague prompts before any file edits.","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Fworkspace-hooks\u002Fskills\u002Fclarify-requirements\u002FSKILL.md",[],[78,79,80,81,82,25,83,84,85,34,86,87,88,19,89],"clarify requirements","clarify","requirements","risk","options","prompts","goal","scanning","context","repo","injection","file",{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":101,"publishedAt":39},"01M6DF79CA4BHJS66J8EZDBPWJ","openmm-md","Run FREE, open-source, Windows-native GPU molecular dynamics via the openmm-md MCP (OpenMM + CUDA on the local NVIDIA GPU). Use for protein \u002F protein-ligand MD when you want a free stack or are on Windows without a Schrodinger license. Covers the full workflow (prepare, solvate, minimize, equilibrate, production, analyze, mmgbsa) AND an honest decision guide vs Schrodinger Desmond, including how to run Desmond legitimately on Linux or a cloud GPU with a real license. This is a deliberately LESSER free alternative and does NOT replace the schrodinger MCP tools. Do NOT use it to pick a free tool when a validated FEP+, WaterMap, IFD-MD, or Prime MM-GBSA result is actually required.","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Fmcp\u002Fopenmm-md-mcp\u002Fskill\u002Fopenmm-md\u002FSKILL.md",[96],{"categoryNo":97,"categoryName":98,"taxonomyType":99,"taxonomyTypeDesc":100},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[102,103,104,36,105,106,107,108,109,110,111,112,113,114,34,115,116,117,118,119],"openmm md","how","decision","linux","analyze","tool","stack","open","NVIDIA","gpu","license","workflow","windows","free","molecular dynamics","local","do","MCP",{"skillNo":121,"name":122,"nameZh":6,"description":123,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":124,"categories":125,"keywords":132,"publishedAt":39},"01M6DF76C1MG85VEWP8VZCBNMN","visual-review","Take screenshots of web pages and UI using Playwright for visual review and iteration","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fvisual-review\u002FSKILL.md",[126,129],{"categoryNo":127,"categoryName":128,"taxonomyType":99,"taxonomyTypeDesc":100},"01JBSEEDCAT000000000000003","演示文稿",{"categoryNo":130,"categoryName":131,"taxonomyType":99,"taxonomyTypeDesc":100},"01JBSEEDCAT000000000000008","测试与质量",[133,134,135,136,137,138,139,140,141,142,103,143,144,145,146],"review","visual","visual review","pages","web","false","iteration","take","check","slides","ui","screenshot","playwright","screenshots",{"skillNo":148,"name":149,"nameZh":6,"description":150,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":151,"categories":152,"keywords":153,"publishedAt":39},"01M6DF73AQVCBZKEN0YXEX9FXP","status","Diagnostic status report — shows metaswarm installation state, project setup, and potential issues","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fstatus\u002FSKILL.md",[],[149,154,155,156,157,158,159,160],"setup","report","project","installation","project setup","issues","state",{"skillNo":162,"name":163,"nameZh":6,"description":164,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":165,"categories":166,"keywords":168,"publishedAt":39},"01M6DF6X6C2XB0P6GX76P7TTYV","start","Use when starting work on any task, when the user mentions metaswarm, or when the user wants to begin tracked development work","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fstart\u002FSKILL.md",[167],{"categoryNo":97,"categoryName":98,"taxonomyType":99,"taxonomyTypeDesc":100},[163,169,25,170,171,172,34,173,174,175,176],"start issue","issue","ready","work on issue","user","work","work on","development",{"skillNo":178,"name":154,"nameZh":6,"description":179,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":180,"categories":181,"keywords":182,"publishedAt":39},"01M6DF6QMJ9AJKFKHHKRT005PQ","Interactive project setup — detects your project, configures metaswarm, writes project-local files","https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Ftools\u002Feditor\u002Fmetaswarm\u002Fskills\u002Fsetup\u002FSKILL.md",[],[154,156,117,183,158,184],"files","interactive"]