[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M74JHARZJ246DYV6SSXFRNXT":3,"skill-same:01M74JHARZJ246DYV6SSXFRNXT":65},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":14,"repositoryHtmlUrl":27,"repositoryHomepageUrl":28,"ownerHtmlUrl":29,"ownerBlogUrl":30,"ownerDescription":31,"updatedAt":32,"install":33,"resources":46},"01M74JHARZJ246DYV6SSXFRNXT","dead-predicate",null,"Use when a hook didn't fire, a reminder never arrived, or before claiming a rule is hook-enforced: a live router rule can match nothing (missing word, JSON escape, shadowing, stale name).","MrBinnacle\u002Fskills","MrBinnacle","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F124333270?v=4","01M74JFP3ZCYZP94CVBY1JR1E6","Matthew Gruber","01M74JFP4K4A75MGTZBC9D836G",1,"MIT","https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fmeta\u002Fdead-predicate\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: dead-predicate\ndescription: \"Use when a hook didn't fire, a reminder never arrived, or before claiming a rule is hook-enforced: a live router rule can match nothing (missing word, JSON escape, shadowing, stale name).\"\n---\n\n# Router skill predicate gap\n\n## Problem\n\nA **router rule** is one entry in a prompt hook's rule file: a skill name and a list of regex\npatterns. The hook (a `UserPromptSubmit` hook in Claude Code) matches each prompt against the\npatterns and, on a match, injects a reminder naming the skill. A skill wired this way looks\nenforced rather than remembered. It is not, when **the hook runs, exits 0, and matches\nnothing**. The wiring is present, the hook is healthy, the config is valid, and nothing\ndistinguishes \"the predicate did not match\" from \"no prompt needed it.\"\n\nThis card probes the live hook and returns, per prompt, `FIRES` or `SILENT`, plus the branch\nthat explains a silence. Origin, 2026-08-18: a rule called MANDATORY before any plan had no\npattern for the word \"plan\" (see § Example). One incident, not a measured rate.\n\n## Use when\n\n- A hook didn't fire, or a reminder never arrived, for a prompt that should have raised it.\n- You are about to write, or repeat, the claim that a discipline is hook-enforced.\n- A rule's patterns were seeded from error signatures or example phrasings and never re-read.\n- The skill did fire once, and you have not checked *which* pattern matched.\n\n## Root cause: five branches\n\n1. **Missing word.** Patterns get seeded from high-precision triggers (error strings, slash\n   commands, distinctive nouns) because the ordinary word looks too broad. The predicate\n   matches specialist phrasing and misses the common request: **model-pull wearing a hook's\n   clothes**.\n2. **JSON escape.** In a JSON string `\"\\b\"` is the backspace character; a regex word boundary\n   is `\"\\\\b\"`. The damaged pattern still compiles and matches nothing. In the maintainer's\n   research repo this happened in six sessions, S312 to S487, and one control missed it by\n   scanning bytes: JSON writes the backspace as two characters, so **decode the JSON before\n   scanning for `\\b`**.\n3. **Shadowing.** A router that stops at the first match lets an earlier rule claim every\n   prompt a later rule's pattern also matches. The later rule is valid and never fires.\n4. **Stale name.** The pattern matches, but the skill it names was renamed or uninstalled, so\n   the reminder points at a name nothing can call. A plugin skill is named\n   `\u003Cplugin>:\u003Cname>` (Claude Code plugin docs, checked 2026-10-05), so promotion into a\n   plugin is a rename.\n5. **Wrong claim, complete predicate.** The rule fires on every case it names, but its advice\n   makes a causal claim and the case that would refute it lies outside the predicate. Every\n   firing looks like confirmation. Write the counter-example and probe whether the rule\n   reaches it.\n\n## Solution\n\n### 1. Test the negative first, against the live hook\n\nProbe the suspect prompt **and a known-good fixture in the same run**. The known-good is a\npositive control; without it the run is uninterpretable:\n\n```sh\nfor p in \"write me a plan for issue 18\" \"\u003Ca phrase you know this rule matches>\"; do\n  out=$(echo \"{\\\"session_id\\\":\\\"neg-$RANDOM$RANDOM\\\",\\\"prompt\\\":\\\"$p\\\"}\" | python \u003Crouter-hook>.py)\n  echo \"$out\" | grep -q \"\u003Cskill-name>\" && echo \"FIRES  : $p\" || echo \"SILENT : $p\"\ndone\n```\n\nEmpty output on the suspect means it did not fire, **but only if the control fired**. A\ncrashed interpreter also prints nothing\n([`vacuous-check`](https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fengineering\u002Fvacuous-check\u002FSKILL.md)\n→ rule 4). Make `session_id` unique per probe: these routers dedupe per session.\n\n### 2 to 4. Scan, read, repair\n\nOpen [probes.md](probes.md) for the scripts. Step 2 decodes the rule file and reports any\npattern holding a control character. Step 3 lists every rule a prompt matches in file order,\nwhich exposes shadowing, and checks each named skill still resolves. Step 4 adds patterns\nthat match the verb plus the noun, so mention stays silent and a request fires.\n\n### 5. Probe positives, then false positives, through the step-1 loop\n\nEvery prompt a user actually types (`I need a plan`) must read `FIRES`. Then run the\nfalse-positive set, **including words that share the stem** (`the plane landed`, `explain\nthe planner architecture`); every line must read `SILENT`.\n\n## Verification\n\nProve the fix by a before\u002Fafter pair on the *same* prompt against the *same* hook:\n\n```\nbefore:  \"write me a plan for issue 18\"  -> silent\nafter:   \"write me a plan for issue 18\"  -> fires\n```\n\nUse the user's own message from the session that exposed the gap; an invented probe can be\naccused of being chosen to fire.\n\n## Example\n\n2026-08-18. A rule file marked `decision-rights` as router-enforced and MANDATORY before any\nhandoff, plan, ADR or subagent prompt, and \"plan\" was in no pattern. The skill had fired that\nsession only because `\\bADR\\b` matched. Three patterns were added; the user's unmatched\nmessage then fired and five false-positive probes stayed silent.\n\n## Notes\n\nOpen [gotchas.md](gotchas.md) when a router rule matches nothing you type, or a green per-rule suite hides a dead pattern. It records each dated case behind the five branches.\n\n- **A passing read is not evidence.** Only piping a prompt into the live hook finds the gap.\n- **Dead wiring is a neighbour.** A hook that never runs, or reads the wrong stdin shape, is a\n  different diagnosis. A hook that runs and names a skill that no longer resolves is branch 4.\n- **The stake is layer placement.** A router rule moves a discipline into the hook layer\n  *only as far as its predicate is complete*. An incomplete one is worse than no hook: it\n  retires the vigilance that would have compensated.\n- **Test per pattern, not per rule.** Give every pattern a fixture only it can satisfy. A\n  green per-rule suite is compatible with any number of dead patterns, because fixtures land\n  on whichever pattern matches first.\n\nVerified against a live `UserPromptSubmit` router hook, 2026-08-18. The stdin fields\n`session_id` and `prompt` match the Claude Code hooks docs, checked 2026-10-05; the per-session\ndedupe belongs to that hook, so re-read yours before assuming it.\n","# Router skill predicate gap\n\n## Problem\n\nA **router rule** is one entry in a prompt hook's rule file: a skill name and a list of regex\npatterns. The hook (a `UserPromptSubmit` hook in Claude Code) matches each prompt against the\npatterns and, on a match, injects a reminder naming the skill. A skill wired this way looks\nenforced rather than remembered. It is not, when **the hook runs, exits 0, and matches\nnothing**. The wiring is present, the hook is healthy, the config is valid, and nothing\ndistinguishes \"the predicate did not match\" from \"no prompt needed it.\"\n\nThis card probes the live hook and returns, per prompt, `FIRES` or `SILENT`, plus the branch\nthat explains a silence. Origin, 2026-08-18: a rule called MANDATORY before any plan had no\npattern for the word \"plan\" (see § Example). One incident, not a measured rate.\n\n## Use when\n\n- A hook didn't fire, or a reminder never arrived, for a prompt that should have raised it.\n- You are about to write, or repeat, the claim that a discipline is hook-enforced.\n- A rule's patterns were seeded from error signatures or example phrasings and never re-read.\n- The skill did fire once, and you have not checked *which* pattern matched.\n\n## Root cause: five branches\n\n1. **Missing word.** Patterns get seeded from high-precision triggers (error strings, slash\n   commands, distinctive nouns) because the ordinary word looks too broad. The predicate\n   matches specialist phrasing and misses the common request: **model-pull wearing a hook's\n   clothes**.\n2. **JSON escape.** In a JSON string `\"\\b\"` is the backspace character; a regex word boundary\n   is `\"\\\\b\"`. The damaged pattern still compiles and matches nothing. In the maintainer's\n   research repo this happened in six sessions, S312 to S487, and one control missed it by\n   scanning bytes: JSON writes the backspace as two characters, so **decode the JSON before\n   scanning for `\\b`**.\n3. **Shadowing.** A router that stops at the first match lets an earlier rule claim every\n   prompt a later rule's pattern also matches. The later rule is valid and never fires.\n4. **Stale name.** The pattern matches, but the skill it names was renamed or uninstalled, so\n   the reminder points at a name nothing can call. A plugin skill is named\n   `\u003Cplugin>:\u003Cname>` (Claude Code plugin docs, checked 2026-10-05), so promotion into a\n   plugin is a rename.\n5. **Wrong claim, complete predicate.** The rule fires on every case it names, but its advice\n   makes a causal claim and the case that would refute it lies outside the predicate. Every\n   firing looks like confirmation. Write the counter-example and probe whether the rule\n   reaches it.\n\n## Solution\n\n### 1. Test the negative first, against the live hook\n\nProbe the suspect prompt **and a known-good fixture in the same run**. The known-good is a\npositive control; without it the run is uninterpretable:\n\n```sh\nfor p in \"write me a plan for issue 18\" \"\u003Ca phrase you know this rule matches>\"; do\n  out=$(echo \"{\\\"session_id\\\":\\\"neg-$RANDOM$RANDOM\\\",\\\"prompt\\\":\\\"$p\\\"}\" | python \u003Crouter-hook>.py)\n  echo \"$out\" | grep -q \"\u003Cskill-name>\" && echo \"FIRES  : $p\" || echo \"SILENT : $p\"\ndone\n```\n\nEmpty output on the suspect means it did not fire, **but only if the control fired**. A\ncrashed interpreter also prints nothing\n([`vacuous-check`](https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fengineering\u002Fvacuous-check\u002FSKILL.md)\n→ rule 4). Make `session_id` unique per probe: these routers dedupe per session.\n\n### 2 to 4. Scan, read, repair\n\nOpen [probes.md](probes.md) for the scripts. Step 2 decodes the rule file and reports any\npattern holding a control character. Step 3 lists every rule a prompt matches in file order,\nwhich exposes shadowing, and checks each named skill still resolves. Step 4 adds patterns\nthat match the verb plus the noun, so mention stays silent and a request fires.\n\n### 5. Probe positives, then false positives, through the step-1 loop\n\nEvery prompt a user actually types (`I need a plan`) must read `FIRES`. Then run the\nfalse-positive set, **including words that share the stem** (`the plane landed`, `explain\nthe planner architecture`); every line must read `SILENT`.\n\n## Verification\n\nProve the fix by a before\u002Fafter pair on the *same* prompt against the *same* hook:\n\n```\nbefore:  \"write me a plan for issue 18\"  -> silent\nafter:   \"write me a plan for issue 18\"  -> fires\n```\n\nUse the user's own message from the session that exposed the gap; an invented probe can be\naccused of being chosen to fire.\n\n## Example\n\n2026-08-18. A rule file marked `decision-rights` as router-enforced and MANDATORY before any\nhandoff, plan, ADR or subagent prompt, and \"plan\" was in no pattern. The skill had fired that\nsession only because `\\bADR\\b` matched. Three patterns were added; the user's unmatched\nmessage then fired and five false-positive probes stayed silent.\n\n## Notes\n\nOpen [gotchas.md](gotchas.md) when a router rule matches nothing you type, or a green per-rule suite hides a dead pattern. It records each dated case behind the five branches.\n\n- **A passing read is not evidence.** Only piping a prompt into the live hook finds the gap.\n- **Dead wiring is a neighbour.** A hook that never runs, or reads the wrong stdin shape, is a\n  different diagnosis. A hook that runs and names a skill that no longer resolves is branch 4.\n- **The stake is layer placement.** A router rule moves a discipline into the hook layer\n  *only as far as its predicate is complete*. An incomplete one is worse than no hook: it\n  retires the vigilance that would have compensated.\n- **Test per pattern, not per rule.** Give every pattern a fixture only it can satisfy. A\n  green per-rule suite is compatible with any number of dead patterns, because fixtures land\n  on whichever pattern matches first.\n\nVerified against a live `UserPromptSubmit` router hook, 2026-08-18. The stdin fields\n`session_id` and `prompt` match the Claude Code hooks docs, checked 2026-10-05; the per-session\ndedupe belongs to that hook, so re-read yours before assuming it.\n",{"name":5,"description":7},"main","fb8ab02da6c849e2cdb704371f5c8c190512d15e","skills\u002Fmeta\u002Fdead-predicate","https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills","https:\u002F\u002Fmrbinnacle.github.io\u002Fskills\u002F","https:\u002F\u002Fgithub.com\u002FMrBinnacle","https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fmatthew-g-83390523","I make AI coding agents work in messy, real environments, and measure whether they did. Ex-Army NCO, ex-J.P. Morgan.","2026-11-15 04:37:29",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add MrBinnacle\u002Fskills --skill dead-predicate",{"tool":39,"command":40},"bunx","bunx skills add MrBinnacle\u002Fskills --skill dead-predicate",{"tool":42,"command":43},"pnpm","pnpm dlx skills add MrBinnacle\u002Fskills --skill dead-predicate","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fmeta\u002Fdead-predicate\u002FSKILL.md",true,[47,53,57,61],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"evals\u002Fevals.json","REFERENCE","参考文档",3220,"https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fmeta\u002Fdead-predicate\u002Fevals\u002Fevals.json",{"path":54,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":55,"sourceHtmlUrl":56},"EVIDENCE.md",6297,"https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fmeta\u002Fdead-predicate\u002FEVIDENCE.md",{"path":58,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":59,"sourceHtmlUrl":60},"gotchas.md",6762,"https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fmeta\u002Fdead-predicate\u002Fgotchas.md",{"path":62,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":63,"sourceHtmlUrl":64},"probes.md",2598,"https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fmeta\u002Fdead-predicate\u002Fprobes.md",[66,73,80,87,94,101],{"skillNo":67,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"publishedAt":19},"01M74JHKQKA5V9N4B4VDAHBGAP","subagent-handback","Use when prompting, briefing or dispatching a research subagent, or when one goes idle with no report. Name a return channel, check its tool grant, verify every claim, negatives first.","https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Forchestration\u002Fsubagent-handback\u002FSKILL.md",[],[],{"skillNo":74,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":77,"categories":78,"keywords":79,"publishedAt":19},"01M74JHGXP4YFAGQCRTMFQ8Q5F","disposition-schema","Use when dispatching parallel agents to verify or adjudicate findings (adversarial verify, security disposition, council fires), or before you consolidate their outputs and risk an ID collision.","https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Forchestration\u002Fdisposition-schema\u002FSKILL.md",[],[],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":19},"01M74JHDMFE7VN27QTC6YRPW22","decision-rights","Use when writing a handoff, plan, ADR or subagent brief, or one says \"do not re-litigate\" or \"final\". Separates evidence access from decision rights and marks what is non-negotiable.","https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Forchestration\u002Fdecision-rights\u002FSKILL.md",[],[],{"skillNo":88,"name":89,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"publishedAt":19},"01M74JH7P66AGN3RHKH7DVA5Y4","vacuous-check","Fail-open success checks. Use when a script says OK but nothing happened, a probe batch returns a false negative or 0 results, or `String(got) === String(want)` passes on a type mismatch.","https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fengineering\u002Fvacuous-check\u002FSKILL.md",[],[],{"skillNo":95,"name":96,"nameZh":6,"description":97,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":98,"categories":99,"keywords":100,"publishedAt":19},"01M74JH59H7FJZW0VR6EXMHQJ5","stale-deploy","Use when asking \"is it live\", a deploy looks stale, or a poll loop exits instantly. Verify a GitHub Pages or CDN deploy serves new content: poll a marker new to this deploy.","https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fengineering\u002Fstale-deploy\u002FSKILL.md",[],[],{"skillNo":102,"name":103,"nameZh":6,"description":104,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":105,"categories":106,"keywords":107,"publishedAt":19},"01M74JH297TR1CZVQHRPBEXGFT","pull-rebase","Use before `git pull` where `pull.rebase` may be `true`, or after a pull rebased your commits. `--no-ff` does not stop the rebase; every local SHA changes. Check config first; recover by reflog.","https:\u002F\u002Fgithub.com\u002FMrBinnacle\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fengineering\u002Fpull-rebase\u002FSKILL.md",[],[]]