[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6DDXP3MVBW36SV1BMEYFG2E":3,"skill-same:01M6DDXP3MVBW36SV1BMEYFG2E":84},{"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},"01M6DDXP3MVBW36SV1BMEYFG2E","claim-verify-gate",null,"Use when an answer, research finding, audit verdict, or status report is about to be treated as true and the next action depends on it. Decomposes the answer into atomic, individually-checkable claims (SAFE, arXiv 2403.18802), runs Chain-of-Verification (arXiv 2309.11495 - draft, plan verification questions, answer them independently, revise), and re-probes each claim with a MECHANICALLY DIFFERENT instrument than produced it (different tool, engine, shell or measured field; re-running the same probe never counts as CONFIRMED). Emits a per-claim CONFIRMED \u002F REFUTED \u002F UNVERIFIED table with the settling experiment, and forces an overclaim to be rewritten to the scope actually measured - an honestly-scoped residual does NOT block. Ships scripts\u002Fclaim_gate.py, a deterministic 10-rule checker with a built-in both-states self-test. NOT for build\u002Ftest\u002Flint claims (hooks\u002Fclaim-guard and verification-before-completion own those), NOT a test runner, NOT a trained critic - no model judge, no new runtime.","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\u002Fskills\u002Fclaim-verify-gate\u002FSKILL.md",[],[19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38],"claim verify gate","verify","claim","gate","hooks","as","runtime","draft","self","deterministic","runner","lint","never","use","model","engine","tool","critic","guard","verification","2026-11-10 10:12:30",false,"---\nname: claim-verify-gate\ndescription: Use when an answer, research finding, audit verdict, or status report is about to be treated as true and the next action depends on it. Decomposes the answer into atomic, individually-checkable claims (SAFE, arXiv 2403.18802), runs Chain-of-Verification (arXiv 2309.11495 - draft, plan verification questions, answer them independently, revise), and re-probes each claim with a MECHANICALLY DIFFERENT instrument than produced it (different tool, engine, shell or measured field; re-running the same probe never counts as CONFIRMED). Emits a per-claim CONFIRMED \u002F REFUTED \u002F UNVERIFIED table with the settling experiment, and forces an overclaim to be rewritten to the scope actually measured - an honestly-scoped residual does NOT block. Ships scripts\u002Fclaim_gate.py, a deterministic 10-rule checker with a built-in both-states self-test. NOT for build\u002Ftest\u002Flint claims (hooks\u002Fclaim-guard and verification-before-completion own those), NOT a test runner, NOT a trained critic - no model judge, no new runtime.\n---\n\n# claim-verify-gate\n\nTurns two rules this estate already has - `AGENTS.md` §7 honesty and\n`rules\u002Fcommon\u002Fsingle-signal-verification.md` - from a habit into a **gate with a\nmechanical half**.\n\nThe premise, in the repo's own words:\n`agent-self\u002Fconfig\u002FAGENTS.md:70` - *\"Sycophancy is in model weights;\nexternal-signal checks are the only reliable counter.\"* A model re-reading its\nown answer is not an external signal. So the parts of this gate that a model\ncould fake are handed to code (`scripts\u002Fclaim_gate.py`), and the model is left\nwith only the part code cannot do: choosing a genuinely different instrument.\n\n## When to run it\n\nRun it when a claim is about to become **load-bearing**:\n\n- a research answer, briefed \"ground truth\", or fan-in digest another agent will build on;\n- an audit \u002F review verdict (\"the hook fires\", \"the config took effect\", \"nothing is enabled\");\n- a status report whose next action is disruptive (a restart, a kill, a re-fix, a rollback);\n- any sentence containing an absolute (`all`, `every`, `zero`, `recall = 1.0`) or a bare hedge\n  (\"should work\", \"likely fine\");\n- before you hand a finding to an orchestrator - `single-signal-verification.md:50`, a\n  subagent's finding is itself a single signal.\n\n## When NOT to run it - route instead\n\nThis gate owns **claims about the world**. It does not own claims about your build.\n\n| The claim is about | Use instead | Why |\n|---|---|---|\n| tests \u002F build \u002F lint \u002F coverage passing | `hooks\u002Fclaim-guard` (automatic) + `verification-before-completion` | already mechanised: a Bash-command ledger cross-referenced per claim category, per turn |\n| whether the TESTS are strong enough | `mutation-check` | probes the tests, not the prose |\n| one value you have leaned on for many steps | `fresh-context-verify` | re-derivation in a clean context, not a second probe |\n| which of N candidate patches to pick | `rubric-verifier` \u002F `execution-grounded-bon` | selection, not verification |\n| how much test surface to run per edit | `two-tier-verify` \u002F `smallest-check-ladder` | scheduling, not truth |\n| whether the DELIVERABLE actually runs | `packaged-artifact-smoke` | needs an artifact, not a table |\n\nIf a claim spans both (an \"it works now\" claim resting on a research premise),\nrun `claim-guard`'s lane for the functional half and this gate for the premise.\n\n## The loop (CoVe, four steps, in this order)\n\nChain-of-Verification (arXiv 2309.11495) is the shape: the model\n(i) drafts, (ii) plans verification questions, (iii) **answers them\nindependently so the answers are not biased by other responses**, (iv) generates\nthe verified response. Step (iii) is the whole point; do not collapse it.\n\n### 1. DECOMPOSE - atomic, individually checkable\n\nSAFE (arXiv 2403.18802) breaks a long-form response into a set of **individual\nfacts** and checks each one separately. Do the same:\n\n- one assertion per claim; if a sentence has two verbs it is two claims;\n- resolve pronouns and \"it\" - each claim must stand alone out of order;\n- separate the **measurement** from the **interpretation** (\"connect took 0.10s\"\n  and \"therefore CDP is fast enough\" are two claims with different evidence);\n- keep the premise: if a recommendation rests on a fact, the fact is its own claim\n  and the recommendation inherits its verdict.\n\nRecord each as a row in the ledger (schema: `references\u002Fclaim-ledger.md`), with\nthe `source_probe` that produced it. Naming the source probe is not bookkeeping -\nstep 3 cannot be checked without it.\n\n### 2. PLAN the verification questions\n\nFor each claim write the question that would **falsify** it, not the one that\nwould confirm it. \"Does `mon` resolve on PATH?\" is confirmable by the broken\ndetector that produced the error; \"which shell was asked, and does a *different*\nshell agree?\" is not.\n\nAsk, per claim: what reading would I see if this were false? If the answer is\n\"the same reading\", the probe is worthless - pick another.\n\n### 3. RE-PROBE independently - the mechanically-different rule\n\n`single-signal-verification.md:34`: the two signals must be mechanically\nindependent - *\"a different shell, a different data field, a different\nmeasurement ... not the same query re-run. Re-running the same probe is not a\nsecond signal.\"*\n\nThe discriminating axes. A verify probe must differ on **at least one**:\n\n| axis | example difference |\n|---|---|\n| `tool` | `curl` vs a Playwright client; arXiv API vs an HTML page fetch |\n| `engine` | one search index vs another; a differ vs an LLM read |\n| `shell` | PowerShell `Get-Command` vs Git-Bash `which` (misses `.cmd` shims) |\n| `field` | process CPU-delta vs file mtime vs byte-count; `whoami` vs `account_list` |\n\n`command` is deliberately **not** an axis. A different command string on the same\ntool, shell and field is the same query re-run - that is exactly the shape\n`fan-out-contract.md:25` measured, where two agents \"CONFIRMED\" a false briefed\nfact by re-running the orchestrator's own broken Git-Bash `which`, evidence tags\nand all.\n\nTwo further rules that carry their own cost history:\n\n- **Verify the detector before trusting a silence** (`single-signal-verification.md:37`).\n  A \"0 results \u002F not found\" reading is often a broken matcher. Confirm the probe\n  can find a known-present item first.\n- **Both-states** (`:40`). Before a probe's silence counts as \"fixed\", run it\n  against the known-BROKEN state and confirm it FIRES there. A probe silent in\n  both states measures nothing. `claim_gate.py --self-test` does exactly this for\n  the gate's own rules; do the same for yours.\n\n### 4. REVISE - and rewrite the overclaim, do not delete it\n\n`fan-out-contract.md:110` measured that **5 of 8** refuted claims were overclaims,\nnot broken work: *\"the code was correct and the sentence was wider than the\nevidence.\"* So the default repair is to the SENTENCE:\n\n- state the scope actually measured, inline;\n- disclose the residual inline, adjacent to the claim - not in a trailing\n  limitations section (`fan-out-contract.md:17`; `:18` measured 40% of real flags\n  parked in the final 15% of their document, where they qualified nothing);\n- name the experiment that would settle the residual;\n- record that the earlier wording was REFUTED. Do not silently delete it, or the\n  dead end gets re-walked.\n\n**A residual that is honestly scoped and disclosed is NOT a refutation and MUST\nNOT block.** A correctly-scoped claim is CONFIRMED. This is the rule that keeps\n\"round N until fully green\" tractable.\n\n## The verdict table\n\nEvery claim ends in exactly one of three states. Ship the table, not prose.\n\n| verdict | what it asserts | mandatory fields |\n|---|---|---|\n| `CONFIRMED` | an independent probe agreed | `verify_probe` differing on >=1 axis, plus >=1 `file:line` or URL anchor |\n| `REFUTED` | an independent probe disagreed | `verify_probe`, plus `revised_text` scoped to what was measured |\n| `UNVERIFIED` | not checked, or not checkable now | `settling_experiment` naming the probe that would decide it |\n\nAttach a Sherman-Kent band to any forward-looking or uncertain claim, and never\nship the band as a bare tag name (`fan-out-contract.md:12`): **almost certain\n90-99 - very likely - likely 55-80 - roughly even 40-60 - unlikely 20-45 - very\nunlikely - remote 1-10 - almost certainly not**.\n\n## The mechanical half - `scripts\u002Fclaim_gate.py`\n\nTen yes\u002Fno questions that a model must not be trusted to answer about its own\noutput. Deterministic, stdlib-only, no network, no model call\n(`fan-out-contract.md:162` - if a question is answerable by an exact algorithm,\nwrite the algorithm).\n\n```\npython .agents\u002Fskills\u002Fclaim-verify-gate\u002Fscripts\u002Fclaim_gate.py --claims ledger.json\npython .agents\u002Fskills\u002Fclaim-verify-gate\u002Fscripts\u002Fclaim_gate.py --claims ledger.json --doc report.md\npython .agents\u002Fskills\u002Fclaim-verify-gate\u002Fscripts\u002Fclaim_gate.py --claims ledger.json --format json\npython .agents\u002Fskills\u002Fclaim-verify-gate\u002Fscripts\u002Fclaim_gate.py --self-test\n```\n\n| rule | fires when |\n|---|---|\n| `no-verify-probe` | CONFIRMED\u002FREFUTED with no `verify_probe` |\n| `same-probe-rerun` | `verify_probe` matches `source_probe` on every axis |\n| `no-evidence-anchor` | CONFIRMED with no `file:line` \u002F URL in `evidence` |\n| `no-settling-experiment` | UNVERIFIED that names no settling experiment |\n| `unscoped-absolute` | absolute wording with no `scope` |\n| `unbanded-hedge` | hedged wording with no band |\n| `unknown-band` | a band outside the enumerated set |\n| `refuted-without-rewrite` | REFUTED with no (or an identical) `revised_text` |\n| `untagged-unverified` | UNVERIFIED claim, no `UNVERIFIED` tag in `--doc` |\n| `trailing-unverified` | every `UNVERIFIED` tag sits in the document's final 15% |\n\nExit `0` gate passed - `1` violations - `2` unusable input. Residuals are counted\nand printed, and never contribute to exit 1.\n\n`--self-test` runs every rule against a known-BROKEN ledger and fails if any rule\nstays silent. Run it before trusting a clean report: a gate that cannot fire is\nnot a gate.\n\n## What this gate does NOT do (inline, so it qualifies something)\n\n- It **never decides whether a claim is true.** It decides whether the ledger is\n  allowed to say so. A ledger can be internally perfect and factually wrong.\n- `trailing-unverified` is a **distribution** check over the whole document, not a\n  per-claim adjacency check: it catches every flag parked in a trailing\n  limitations section, and does NOT prove any individual tag sits next to its own\n  sentence. Settling experiment: extend the ledger with a character offset per\n  claim and assert the tag lands within N characters of it.\n- Probe independence is checked on **declared** axes. An author who mislabels two\n  runs of one tool as different tools defeats it. It raises the cost of faking\n  from zero to a deliberate lie; it does not make faking impossible.\n- The absolute\u002Fhedge word lists are heuristics tuned on this estate's measured\n  overclaim shapes. They will miss novel phrasings. UNVERIFIED (likely 55-80) that\n  the lists generalise to another corpus; settling experiment: run the gate over\n  `harness-sota-2026-07-18-corpus\u002F` and score flagged-vs-actual overclaims.\n\n## Composes with\n\n- `rules\u002Fcommon\u002Fsingle-signal-verification.md` - the rule this enforces; read it, this does not restate it.\n- `rules\u002Fcommon\u002Ffan-out-contract.md` §A2\u002FA3\u002FA4, §B3, §C2 - the fan-out and fan-in obligations the table satisfies.\n- `agent-self\u002Fconfig\u002FAGENTS.md` §4 terminal states, §7 honesty.\n- `hooks\u002Fclaim-guard\u002F` - the automatic per-turn lane for build\u002Ftest claims.\n- `mutation-check`, `fresh-context-verify`, `simplest-probe-first` - sibling probes, different questions.\n\n## Sources\n\n- Wei et al., *Long-form factuality in large language models* (SAFE, LongFact, F1@K),\n  arXiv:2403.18802 - decomposition into individual facts, each rated separately.\n- Dhuliawala et al., *Chain-of-Verification Reduces Hallucination in Large Language Models*,\n  arXiv:2309.11495 - draft, plan verification questions, answer independently, verified response.\n","# claim-verify-gate\n\nTurns two rules this estate already has - `AGENTS.md` §7 honesty and\n`rules\u002Fcommon\u002Fsingle-signal-verification.md` - from a habit into a **gate with a\nmechanical half**.\n\nThe premise, in the repo's own words:\n`agent-self\u002Fconfig\u002FAGENTS.md:70` - *\"Sycophancy is in model weights;\nexternal-signal checks are the only reliable counter.\"* A model re-reading its\nown answer is not an external signal. So the parts of this gate that a model\ncould fake are handed to code (`scripts\u002Fclaim_gate.py`), and the model is left\nwith only the part code cannot do: choosing a genuinely different instrument.\n\n## When to run it\n\nRun it when a claim is about to become **load-bearing**:\n\n- a research answer, briefed \"ground truth\", or fan-in digest another agent will build on;\n- an audit \u002F review verdict (\"the hook fires\", \"the config took effect\", \"nothing is enabled\");\n- a status report whose next action is disruptive (a restart, a kill, a re-fix, a rollback);\n- any sentence containing an absolute (`all`, `every`, `zero`, `recall = 1.0`) or a bare hedge\n  (\"should work\", \"likely fine\");\n- before you hand a finding to an orchestrator - `single-signal-verification.md:50`, a\n  subagent's finding is itself a single signal.\n\n## When NOT to run it - route instead\n\nThis gate owns **claims about the world**. It does not own claims about your build.\n\n| The claim is about | Use instead | Why |\n|---|---|---|\n| tests \u002F build \u002F lint \u002F coverage passing | `hooks\u002Fclaim-guard` (automatic) + `verification-before-completion` | already mechanised: a Bash-command ledger cross-referenced per claim category, per turn |\n| whether the TESTS are strong enough | `mutation-check` | probes the tests, not the prose |\n| one value you have leaned on for many steps | `fresh-context-verify` | re-derivation in a clean context, not a second probe |\n| which of N candidate patches to pick | `rubric-verifier` \u002F `execution-grounded-bon` | selection, not verification |\n| how much test surface to run per edit | `two-tier-verify` \u002F `smallest-check-ladder` | scheduling, not truth |\n| whether the DELIVERABLE actually runs | `packaged-artifact-smoke` | needs an artifact, not a table |\n\nIf a claim spans both (an \"it works now\" claim resting on a research premise),\nrun `claim-guard`'s lane for the functional half and this gate for the premise.\n\n## The loop (CoVe, four steps, in this order)\n\nChain-of-Verification (arXiv 2309.11495) is the shape: the model\n(i) drafts, (ii) plans verification questions, (iii) **answers them\nindependently so the answers are not biased by other responses**, (iv) generates\nthe verified response. Step (iii) is the whole point; do not collapse it.\n\n### 1. DECOMPOSE - atomic, individually checkable\n\nSAFE (arXiv 2403.18802) breaks a long-form response into a set of **individual\nfacts** and checks each one separately. Do the same:\n\n- one assertion per claim; if a sentence has two verbs it is two claims;\n- resolve pronouns and \"it\" - each claim must stand alone out of order;\n- separate the **measurement** from the **interpretation** (\"connect took 0.10s\"\n  and \"therefore CDP is fast enough\" are two claims with different evidence);\n- keep the premise: if a recommendation rests on a fact, the fact is its own claim\n  and the recommendation inherits its verdict.\n\nRecord each as a row in the ledger (schema: `references\u002Fclaim-ledger.md`), with\nthe `source_probe` that produced it. Naming the source probe is not bookkeeping -\nstep 3 cannot be checked without it.\n\n### 2. PLAN the verification questions\n\nFor each claim write the question that would **falsify** it, not the one that\nwould confirm it. \"Does `mon` resolve on PATH?\" is confirmable by the broken\ndetector that produced the error; \"which shell was asked, and does a *different*\nshell agree?\" is not.\n\nAsk, per claim: what reading would I see if this were false? If the answer is\n\"the same reading\", the probe is worthless - pick another.\n\n### 3. RE-PROBE independently - the mechanically-different rule\n\n`single-signal-verification.md:34`: the two signals must be mechanically\nindependent - *\"a different shell, a different data field, a different\nmeasurement ... not the same query re-run. Re-running the same probe is not a\nsecond signal.\"*\n\nThe discriminating axes. A verify probe must differ on **at least one**:\n\n| axis | example difference |\n|---|---|\n| `tool` | `curl` vs a Playwright client; arXiv API vs an HTML page fetch |\n| `engine` | one search index vs another; a differ vs an LLM read |\n| `shell` | PowerShell `Get-Command` vs Git-Bash `which` (misses `.cmd` shims) |\n| `field` | process CPU-delta vs file mtime vs byte-count; `whoami` vs `account_list` |\n\n`command` is deliberately **not** an axis. A different command string on the same\ntool, shell and field is the same query re-run - that is exactly the shape\n`fan-out-contract.md:25` measured, where two agents \"CONFIRMED\" a false briefed\nfact by re-running the orchestrator's own broken Git-Bash `which`, evidence tags\nand all.\n\nTwo further rules that carry their own cost history:\n\n- **Verify the detector before trusting a silence** (`single-signal-verification.md:37`).\n  A \"0 results \u002F not found\" reading is often a broken matcher. Confirm the probe\n  can find a known-present item first.\n- **Both-states** (`:40`). Before a probe's silence counts as \"fixed\", run it\n  against the known-BROKEN state and confirm it FIRES there. A probe silent in\n  both states measures nothing. `claim_gate.py --self-test` does exactly this for\n  the gate's own rules; do the same for yours.\n\n### 4. REVISE - and rewrite the overclaim, do not delete it\n\n`fan-out-contract.md:110` measured that **5 of 8** refuted claims were overclaims,\nnot broken work: *\"the code was correct and the sentence was wider than the\nevidence.\"* So the default repair is to the SENTENCE:\n\n- state the scope actually measured, inline;\n- disclose the residual inline, adjacent to the claim - not in a trailing\n  limitations section (`fan-out-contract.md:17`; `:18` measured 40% of real flags\n  parked in the final 15% of their document, where they qualified nothing);\n- name the experiment that would settle the residual;\n- record that the earlier wording was REFUTED. Do not silently delete it, or the\n  dead end gets re-walked.\n\n**A residual that is honestly scoped and disclosed is NOT a refutation and MUST\nNOT block.** A correctly-scoped claim is CONFIRMED. This is the rule that keeps\n\"round N until fully green\" tractable.\n\n## The verdict table\n\nEvery claim ends in exactly one of three states. Ship the table, not prose.\n\n| verdict | what it asserts | mandatory fields |\n|---|---|---|\n| `CONFIRMED` | an independent probe agreed | `verify_probe` differing on >=1 axis, plus >=1 `file:line` or URL anchor |\n| `REFUTED` | an independent probe disagreed | `verify_probe`, plus `revised_text` scoped to what was measured |\n| `UNVERIFIED` | not checked, or not checkable now | `settling_experiment` naming the probe that would decide it |\n\nAttach a Sherman-Kent band to any forward-looking or uncertain claim, and never\nship the band as a bare tag name (`fan-out-contract.md:12`): **almost certain\n90-99 - very likely - likely 55-80 - roughly even 40-60 - unlikely 20-45 - very\nunlikely - remote 1-10 - almost certainly not**.\n\n## The mechanical half - `scripts\u002Fclaim_gate.py`\n\nTen yes\u002Fno questions that a model must not be trusted to answer about its own\noutput. Deterministic, stdlib-only, no network, no model call\n(`fan-out-contract.md:162` - if a question is answerable by an exact algorithm,\nwrite the algorithm).\n\n```\npython .agents\u002Fskills\u002Fclaim-verify-gate\u002Fscripts\u002Fclaim_gate.py --claims ledger.json\npython .agents\u002Fskills\u002Fclaim-verify-gate\u002Fscripts\u002Fclaim_gate.py --claims ledger.json --doc report.md\npython .agents\u002Fskills\u002Fclaim-verify-gate\u002Fscripts\u002Fclaim_gate.py --claims ledger.json --format json\npython .agents\u002Fskills\u002Fclaim-verify-gate\u002Fscripts\u002Fclaim_gate.py --self-test\n```\n\n| rule | fires when |\n|---|---|\n| `no-verify-probe` | CONFIRMED\u002FREFUTED with no `verify_probe` |\n| `same-probe-rerun` | `verify_probe` matches `source_probe` on every axis |\n| `no-evidence-anchor` | CONFIRMED with no `file:line` \u002F URL in `evidence` |\n| `no-settling-experiment` | UNVERIFIED that names no settling experiment |\n| `unscoped-absolute` | absolute wording with no `scope` |\n| `unbanded-hedge` | hedged wording with no band |\n| `unknown-band` | a band outside the enumerated set |\n| `refuted-without-rewrite` | REFUTED with no (or an identical) `revised_text` |\n| `untagged-unverified` | UNVERIFIED claim, no `UNVERIFIED` tag in `--doc` |\n| `trailing-unverified` | every `UNVERIFIED` tag sits in the document's final 15% |\n\nExit `0` gate passed - `1` violations - `2` unusable input. Residuals are counted\nand printed, and never contribute to exit 1.\n\n`--self-test` runs every rule against a known-BROKEN ledger and fails if any rule\nstays silent. Run it before trusting a clean report: a gate that cannot fire is\nnot a gate.\n\n## What this gate does NOT do (inline, so it qualifies something)\n\n- It **never decides whether a claim is true.** It decides whether the ledger is\n  allowed to say so. A ledger can be internally perfect and factually wrong.\n- `trailing-unverified` is a **distribution** check over the whole document, not a\n  per-claim adjacency check: it catches every flag parked in a trailing\n  limitations section, and does NOT prove any individual tag sits next to its own\n  sentence. Settling experiment: extend the ledger with a character offset per\n  claim and assert the tag lands within N characters of it.\n- Probe independence is checked on **declared** axes. An author who mislabels two\n  runs of one tool as different tools defeats it. It raises the cost of faking\n  from zero to a deliberate lie; it does not make faking impossible.\n- The absolute\u002Fhedge word lists are heuristics tuned on this estate's measured\n  overclaim shapes. They will miss novel phrasings. UNVERIFIED (likely 55-80) that\n  the lists generalise to another corpus; settling experiment: run the gate over\n  `harness-sota-2026-07-18-corpus\u002F` and score flagged-vs-actual overclaims.\n\n## Composes with\n\n- `rules\u002Fcommon\u002Fsingle-signal-verification.md` - the rule this enforces; read it, this does not restate it.\n- `rules\u002Fcommon\u002Ffan-out-contract.md` §A2\u002FA3\u002FA4, §B3, §C2 - the fan-out and fan-in obligations the table satisfies.\n- `agent-self\u002Fconfig\u002FAGENTS.md` §4 terminal states, §7 honesty.\n- `hooks\u002Fclaim-guard\u002F` - the automatic per-turn lane for build\u002Ftest claims.\n- `mutation-check`, `fresh-context-verify`, `simplest-probe-first` - sibling probes, different questions.\n\n## Sources\n\n- Wei et al., *Long-form factuality in large language models* (SAFE, LongFact, F1@K),\n  arXiv:2403.18802 - decomposition into individual facts, each rated separately.\n- Dhuliawala et al., *Chain-of-Verification Reduces Hallucination in Large Language Models*,\n  arXiv:2309.11495 - draft, plan verification questions, answer independently, verified response.\n",{"name":5,"description":7},"main","a298c929b6cbd662e1fce54302ce520d5ceea1a5","skills\u002Fclaim-verify-gate","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 claim-verify-gate",{"tool":56,"command":57},"bunx","bunx skills add Prekzursil\u002Fagent-skills-toolchain --skill claim-verify-gate",{"tool":59,"command":60},"pnpm","pnpm dlx skills add Prekzursil\u002Fagent-skills-toolchain --skill claim-verify-gate","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fclaim-verify-gate\u002FSKILL.md",true,[64,70,76,80],{"path":65,"resourceType":66,"resourceTypeDesc":67,"sizeBytes":68,"sourceHtmlUrl":69},"references\u002Fclaim-ledger.md","REFERENCE","参考文档",3706,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fclaim-verify-gate\u002Freferences\u002Fclaim-ledger.md",{"path":71,"resourceType":72,"resourceTypeDesc":73,"sizeBytes":74,"sourceHtmlUrl":75},"scripts\u002Fclaim_gate.py","SCRIPT","脚本",26243,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fclaim-verify-gate\u002Fscripts\u002Fclaim_gate.py",{"path":77,"resourceType":72,"resourceTypeDesc":73,"sizeBytes":78,"sourceHtmlUrl":79},"scripts\u002Ftest_claim_gate_mutations.py",4200,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fclaim-verify-gate\u002Fscripts\u002Ftest_claim_gate_mutations.py",{"path":81,"resourceType":72,"resourceTypeDesc":73,"sizeBytes":82,"sourceHtmlUrl":83},"scripts\u002Ftest_claim_gate.py",21551,"https:\u002F\u002Fgithub.com\u002FPrekzursil\u002Fagent-skills-toolchain\u002Fblob\u002Fmain\u002Fskills\u002Fclaim-verify-gate\u002Fscripts\u002Ftest_claim_gate.py",[85,106,136,163,177,193],{"skillNo":86,"name":87,"nameZh":6,"description":88,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":89,"categories":90,"keywords":91,"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",[],[92,93,94,95,96,97,98,99,100,32,101,102,103,104,105],"clarify requirements","clarify","requirements","risk","options","task","prompts","goal","scanning","context","repo","injection","execution","file",{"skillNo":107,"name":108,"nameZh":6,"description":109,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":110,"categories":111,"keywords":117,"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",[112],{"categoryNo":113,"categoryName":114,"taxonomyType":115,"taxonomyTypeDesc":116},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[118,119,120,121,122,123,35,124,125,126,127,128,129,130,32,131,132,133,134,135],"openmm md","how","decision","pick","linux","analyze","stack","open","NVIDIA","gpu","license","workflow","windows","free","molecular dynamics","local","do","MCP",{"skillNo":137,"name":138,"nameZh":6,"description":139,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":140,"categories":141,"keywords":148,"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",[142,145],{"categoryNo":143,"categoryName":144,"taxonomyType":115,"taxonomyTypeDesc":116},"01JBSEEDCAT000000000000003","演示文稿",{"categoryNo":146,"categoryName":147,"taxonomyType":115,"taxonomyTypeDesc":116},"01JBSEEDCAT000000000000008","测试与质量",[149,150,151,152,153,154,155,156,157,158,119,159,160,161,162],"review","visual","visual review","pages","web","false","iteration","take","check","slides","ui","screenshot","playwright","screenshots",{"skillNo":164,"name":165,"nameZh":6,"description":166,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":167,"categories":168,"keywords":169,"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",[],[165,170,171,172,173,174,175,176],"setup","report","project","installation","project setup","issues","state",{"skillNo":178,"name":179,"nameZh":6,"description":180,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":181,"categories":182,"keywords":184,"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",[183],{"categoryNo":113,"categoryName":114,"taxonomyType":115,"taxonomyTypeDesc":116},[179,185,97,186,187,188,32,189,190,191,192],"start issue","issue","ready","work on issue","user","work","work on","development",{"skillNo":194,"name":170,"nameZh":6,"description":195,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":196,"categories":197,"keywords":198,"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",[],[170,172,133,199,174,200],"files","interactive"]