[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77XCCB6Z2G008Q9XG3XR629":3,"skill-same:01M77XCCB6Z2G008Q9XG3XR629":67},{"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":6,"ownerHtmlUrl":28,"ownerBlogUrl":29,"ownerDescription":6,"updatedAt":30,"install":31,"resources":44},"01M77XCCB6Z2G008Q9XG3XR629","distill-domain-knowledge",null,"Use when capturing a domain expert's tacit knowledge — domain rules, decision rationale, edge cases, system logic — into reusable, verified docs and triggerable skills. Use when someone wants to \"write down what's in my head\" so future AI sessions get the domain right, or when building a project glossary \u002F ubiquitous language to remove term ambiguity.","YunxiangHuang\u002Fdistill-me","YunxiangHuang","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F7898555?v=4","01M77XCANY73RWK3XA2YPYW5YW","Yunxiang Huang","01M77XCAPNFV2K4KXTB88W238F",0,"MIT","https:\u002F\u002Fgithub.com\u002FYunxiangHuang\u002Fdistill-me\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdistill-domain-knowledge\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: distill-domain-knowledge\ndescription: Use when capturing a domain expert's tacit knowledge — domain rules, decision rationale, edge cases, system logic — into reusable, verified docs and triggerable skills. Use when someone wants to \"write down what's in my head\" so future AI sessions get the domain right, or when building a project glossary \u002F ubiquitous language to remove term ambiguity.\n---\n\n# Distilling Domain Knowledge\n\nTurn an expert's tacit knowledge into **ground-truth-verified** docs + a triggerable skill, with a living glossary that removes term ambiguity.\n\n## Core principle — two failure modes, two pillars\n\n- **Content drift** (memory ≠ reality) → **Verify before capture**: the expert's spoken \"how-it-works\" claims are DEFAULT-UNVERIFIED until checked against ground truth.\n- **Reference drift** (one word → many meanings) → **Ubiquitous-language glossary**: every term gets ONE canonical definition; overloaded words get split\u002Fdisambiguated.\n\nA polished knowledge base full of unverified or ambiguous claims is worse than none — it confidently misleads every future session.\n\n- **Content before container** (a *format*-layer failure mode): don't author a skill-format spec \u002F directory scheme before extracting real content — the dominant failure is a polished empty shell (perfect format, nothing externalized). Markdown source-of-truth comes first; prove the protocol on ONE slice end-to-end (a walking skeleton) before scaling the format out. Value priority: **content > container**.\n\n## When to use\n- An expert wants to externalize domain knowledge \u002F decision rationale \u002F logic.\n- Documenting a subsystem so future AI sessions implement\u002Freview it correctly.\n- Term ambiguity keeps causing confusion → build\u002Fmaintain the glossary.\n\n## Method\n\nSub-tools: **grill-me** for the interview (step ②, **mandatory** — external dependency from `github.com\u002Fmattpocock\u002Fskills`, install it to run the Grill step by the book), **writing-skills** for the final skill, the **Workflow tool \u002F parallel agents** for verification fan-out. Degrade gracefully only if a sub-tool genuinely doesn't exist.\n\n```\n① Scaffold → ② Grill → ③ Verify → ④ Capture(verified only) → ⑤ Glossary final-align → ⑥ Distill\n   ├──────────── Glossary-first term admission：贯穿全程，新名词出现的瞬间执行 ────────────┤\n   └──────── Memory loop（有记忆库则必走）：recall@① · re-verify@③ · index@④ · reconcile@⑥ ────────┘\n```\n\n**This SKILL.md is the spine.** Each step's detailed discipline lives in a reference file — open it when you reach that step (don't try to run the step from the one-liner alone; the traps are in the detail):\n\n| Step \u002F cross-cut | Read |\n|---|---|\n| ② Grill (the relentless interview + its traps) | `references\u002Fgrilling.md` |\n| ③ Verify + ④ Capture (the gate; code-vs-prod; bug-claim reality check) | `references\u002Fverify-gate.md` |\n| Glossary-first admission (runs throughout) + ⑤ final-align | `references\u002Fglossary-first.md` |\n| Memory loop (runs throughout, if a store exists) | `references\u002Fmemory-loop.md` |\n| ⑥ Distill — RED→GREEN skill validation | `references\u002Fred-green-validation.md` |\n\n### Cross-cutting disciplines — apply at EVERY step, not only where numbered\n\nA term surfaces mid-Verify; an unexercised enum is classified at Capture; the store is touched at four steps. These don't belong to one step — keep their CORE in hand throughout (the per-step references give depth, but you must apply these even when you never open that reference):\n\n- **Glossary-first admission** — the moment ANY new term surfaces (the expert's words AND names *you* coin: files, skills, domain labels): duplicate-check → org-collision-check (**ask the expert**) → admit with the **headword in its original language** (a translated\u002Fcoined umbrella silently merges distinct concepts — e.g. one label for both `refund` and `payout`). → `references\u002Fglossary-first.md`\n- **Memory loop** (if a store exists) — recall@① · re-verify recalled entries@③ · index on capture (**delete the stale entry first**)@④ · reconcile@⑥. A slice that leaves stale entries behind is not done. → `references\u002Fmemory-loop.md`\n- **Never-produced artifact → don't self-label** — a state never written \u002F flag never set \u002F field nothing populates is forward-reserved *vs* backward-residual ambiguous; only the expert (or a `\u002F\u002F DEPRECATED` comment \u002F migration) decides — default `pending` → ask, don't pick \"dead\" or \"reserved\". A *narrowly*-produced one (admin-tool-only) is **live-but-narrow**, not this trap. → `references\u002Fgrilling.md`\n- **Honesty over fluency** — no recorded reason? capture exactly that; never backfill a plausible \"why\" (future sessions inherit it as fact). Applies to recommended answers (Grill) AND to what you write (Capture).\n- **Verify before capture** — every mechanical \u002F how-it-works claim is default-unverified, and code-verified ≠ prod-verified. → `references\u002Fverify-gate.md`\n- **Sub-agent fan-out discipline** — when dispatching agents for code reconnaissance\u002Fjudgement (Scaffold enumeration, Verify falsification, RED→GREEN), two non-negotiables: ① **model** — judgement-class fan-out runs on a production-grade model, NOT the lightweight default; a weak model fails by being dim, not by lacking info, so its coverage call \u002F verdict is suspect. ② **scan depth** — each agent MUST read the actual code to judge coverage\u002Fmechanism; NEVER infer from directory layout or file names alone (a shallow dir-skim makes the coverage call unreliable). Require a `scannedSummary` back, proving it covered the assigned set rather than guessed from names.\n\n### The six steps (essence — detail in the reference files)\n\n1. **Scaffold (light, one pass).** Build a \"current understanding\" skeleton from existing ground truth — code, schemas, docs, data, dashboards, tickets\u002FPRs, prior memory (with a store, this is Memory loop *recall*). Emit a **typed entity catalog** as the spine: classify each entity as **module \u002F cross-cutting function \u002F config domain**, one-line definition + **upstream→downstream**. Just enough to grill against; **do NOT deep-dive** — and a *thorough* Scaffold raises, not lowers, how much grilling is left (see `references\u002Fgrilling.md`). Reserve parallel agent\u002FWorkflow fan-out for **Verify** (breadth-checking mechanical claims against ground truth), NOT for mass-re-reading code at Scaffold — the high-value work is the serial, human-in-loop grill.\n2. **Grill — invoke `grill-me` (MANDATORY, don't hand-roll it).** Interview relentlessly, ONE branch at a time, each question carrying YOUR recommended answer, until each branch bottoms out. The traps (scaffold-induced laziness, manufacturing reasons, forward-reserved vs backward-residual) are in `references\u002Fgrilling.md` — read it.\n3. **Verify — the gate.** Every mechanical\u002Ffactual\u002Fhow-it-works claim is UNVERIFIED; proactively go find ground truth (fan out agents\u002FWorkflow). \"Why\u002Fintent\" is the expert's authority. On discrepancy, **re-grill** with the evidence. End with a `pending\u002Fgaps.md` punch-list. Code-vs-prod limits, the bug-claim reality check, and the answer-mismatch signal are in `references\u002Fverify-gate.md`.\n4. **Capture (verified only).** Small markdown files, one concept each, tagged `✅ verified @ path:line` (anchor REQUIRED), `⚠️ expert-asserted`, or `pending` (→ punch-list, out of the verified body). Then Memory loop *index*. Detail in `references\u002Fverify-gate.md`.\n5. **Glossary final-align.** ONE living glossary; every term → one canonical definition + scope-type, **headword in the term's original language**. Sweep for terms that slipped admission. Detail in `references\u002Fglossary-first.md`.\n6. **Distill.** Generate a focused **triggerable skill** (and\u002For ADR\u002Frunbook) from the VERIFIED knowledge; test it **RED→GREEN** per `references\u002Fred-green-validation.md`. Close with Memory loop *reconcile* — a slice that leaves stale store entries behind is NOT done.\n\n## Output shape (per domain slice)\n- `knowledge\u002F\u003Carea>\u002F\u003Cconcept>.md` — verified source of truth; every ✅ claim carries a `path:line` code anchor.\n- `\u003Cglossary>.md` — the living ubiquitous language (each term: canonical definition + scope-type, original-language headword).\n- `pending\u002Fgaps.md` — punch-list of unresolved \u002F unverifiable items (claim · why · what's-needed · owner).\n- `.claude\u002Fskills\u002F\u003Cname>\u002FSKILL.md` — distilled, RED→GREEN-verified skill (symlink to `~\u002F.claude\u002Fskills\u002F` to make it live everywhere).\n- **Memory-store delta** (whenever a store exists) — every new\u002Frevised file (re)indexed with stale entries deleted first; every recalled\u002Ftouched\u002Fsuperseded entry reconciled; verified byproducts landed as point-in-time-verified entries.\n","# Distilling Domain Knowledge\n\nTurn an expert's tacit knowledge into **ground-truth-verified** docs + a triggerable skill, with a living glossary that removes term ambiguity.\n\n## Core principle — two failure modes, two pillars\n\n- **Content drift** (memory ≠ reality) → **Verify before capture**: the expert's spoken \"how-it-works\" claims are DEFAULT-UNVERIFIED until checked against ground truth.\n- **Reference drift** (one word → many meanings) → **Ubiquitous-language glossary**: every term gets ONE canonical definition; overloaded words get split\u002Fdisambiguated.\n\nA polished knowledge base full of unverified or ambiguous claims is worse than none — it confidently misleads every future session.\n\n- **Content before container** (a *format*-layer failure mode): don't author a skill-format spec \u002F directory scheme before extracting real content — the dominant failure is a polished empty shell (perfect format, nothing externalized). Markdown source-of-truth comes first; prove the protocol on ONE slice end-to-end (a walking skeleton) before scaling the format out. Value priority: **content > container**.\n\n## When to use\n- An expert wants to externalize domain knowledge \u002F decision rationale \u002F logic.\n- Documenting a subsystem so future AI sessions implement\u002Freview it correctly.\n- Term ambiguity keeps causing confusion → build\u002Fmaintain the glossary.\n\n## Method\n\nSub-tools: **grill-me** for the interview (step ②, **mandatory** — external dependency from `github.com\u002Fmattpocock\u002Fskills`, install it to run the Grill step by the book), **writing-skills** for the final skill, the **Workflow tool \u002F parallel agents** for verification fan-out. Degrade gracefully only if a sub-tool genuinely doesn't exist.\n\n```\n① Scaffold → ② Grill → ③ Verify → ④ Capture(verified only) → ⑤ Glossary final-align → ⑥ Distill\n   ├──────────── Glossary-first term admission：贯穿全程，新名词出现的瞬间执行 ────────────┤\n   └──────── Memory loop（有记忆库则必走）：recall@① · re-verify@③ · index@④ · reconcile@⑥ ────────┘\n```\n\n**This SKILL.md is the spine.** Each step's detailed discipline lives in a reference file — open it when you reach that step (don't try to run the step from the one-liner alone; the traps are in the detail):\n\n| Step \u002F cross-cut | Read |\n|---|---|\n| ② Grill (the relentless interview + its traps) | `references\u002Fgrilling.md` |\n| ③ Verify + ④ Capture (the gate; code-vs-prod; bug-claim reality check) | `references\u002Fverify-gate.md` |\n| Glossary-first admission (runs throughout) + ⑤ final-align | `references\u002Fglossary-first.md` |\n| Memory loop (runs throughout, if a store exists) | `references\u002Fmemory-loop.md` |\n| ⑥ Distill — RED→GREEN skill validation | `references\u002Fred-green-validation.md` |\n\n### Cross-cutting disciplines — apply at EVERY step, not only where numbered\n\nA term surfaces mid-Verify; an unexercised enum is classified at Capture; the store is touched at four steps. These don't belong to one step — keep their CORE in hand throughout (the per-step references give depth, but you must apply these even when you never open that reference):\n\n- **Glossary-first admission** — the moment ANY new term surfaces (the expert's words AND names *you* coin: files, skills, domain labels): duplicate-check → org-collision-check (**ask the expert**) → admit with the **headword in its original language** (a translated\u002Fcoined umbrella silently merges distinct concepts — e.g. one label for both `refund` and `payout`). → `references\u002Fglossary-first.md`\n- **Memory loop** (if a store exists) — recall@① · re-verify recalled entries@③ · index on capture (**delete the stale entry first**)@④ · reconcile@⑥. A slice that leaves stale entries behind is not done. → `references\u002Fmemory-loop.md`\n- **Never-produced artifact → don't self-label** — a state never written \u002F flag never set \u002F field nothing populates is forward-reserved *vs* backward-residual ambiguous; only the expert (or a `\u002F\u002F DEPRECATED` comment \u002F migration) decides — default `pending` → ask, don't pick \"dead\" or \"reserved\". A *narrowly*-produced one (admin-tool-only) is **live-but-narrow**, not this trap. → `references\u002Fgrilling.md`\n- **Honesty over fluency** — no recorded reason? capture exactly that; never backfill a plausible \"why\" (future sessions inherit it as fact). Applies to recommended answers (Grill) AND to what you write (Capture).\n- **Verify before capture** — every mechanical \u002F how-it-works claim is default-unverified, and code-verified ≠ prod-verified. → `references\u002Fverify-gate.md`\n- **Sub-agent fan-out discipline** — when dispatching agents for code reconnaissance\u002Fjudgement (Scaffold enumeration, Verify falsification, RED→GREEN), two non-negotiables: ① **model** — judgement-class fan-out runs on a production-grade model, NOT the lightweight default; a weak model fails by being dim, not by lacking info, so its coverage call \u002F verdict is suspect. ② **scan depth** — each agent MUST read the actual code to judge coverage\u002Fmechanism; NEVER infer from directory layout or file names alone (a shallow dir-skim makes the coverage call unreliable). Require a `scannedSummary` back, proving it covered the assigned set rather than guessed from names.\n\n### The six steps (essence — detail in the reference files)\n\n1. **Scaffold (light, one pass).** Build a \"current understanding\" skeleton from existing ground truth — code, schemas, docs, data, dashboards, tickets\u002FPRs, prior memory (with a store, this is Memory loop *recall*). Emit a **typed entity catalog** as the spine: classify each entity as **module \u002F cross-cutting function \u002F config domain**, one-line definition + **upstream→downstream**. Just enough to grill against; **do NOT deep-dive** — and a *thorough* Scaffold raises, not lowers, how much grilling is left (see `references\u002Fgrilling.md`). Reserve parallel agent\u002FWorkflow fan-out for **Verify** (breadth-checking mechanical claims against ground truth), NOT for mass-re-reading code at Scaffold — the high-value work is the serial, human-in-loop grill.\n2. **Grill — invoke `grill-me` (MANDATORY, don't hand-roll it).** Interview relentlessly, ONE branch at a time, each question carrying YOUR recommended answer, until each branch bottoms out. The traps (scaffold-induced laziness, manufacturing reasons, forward-reserved vs backward-residual) are in `references\u002Fgrilling.md` — read it.\n3. **Verify — the gate.** Every mechanical\u002Ffactual\u002Fhow-it-works claim is UNVERIFIED; proactively go find ground truth (fan out agents\u002FWorkflow). \"Why\u002Fintent\" is the expert's authority. On discrepancy, **re-grill** with the evidence. End with a `pending\u002Fgaps.md` punch-list. Code-vs-prod limits, the bug-claim reality check, and the answer-mismatch signal are in `references\u002Fverify-gate.md`.\n4. **Capture (verified only).** Small markdown files, one concept each, tagged `✅ verified @ path:line` (anchor REQUIRED), `⚠️ expert-asserted`, or `pending` (→ punch-list, out of the verified body). Then Memory loop *index*. Detail in `references\u002Fverify-gate.md`.\n5. **Glossary final-align.** ONE living glossary; every term → one canonical definition + scope-type, **headword in the term's original language**. Sweep for terms that slipped admission. Detail in `references\u002Fglossary-first.md`.\n6. **Distill.** Generate a focused **triggerable skill** (and\u002For ADR\u002Frunbook) from the VERIFIED knowledge; test it **RED→GREEN** per `references\u002Fred-green-validation.md`. Close with Memory loop *reconcile* — a slice that leaves stale store entries behind is NOT done.\n\n## Output shape (per domain slice)\n- `knowledge\u002F\u003Carea>\u002F\u003Cconcept>.md` — verified source of truth; every ✅ claim carries a `path:line` code anchor.\n- `\u003Cglossary>.md` — the living ubiquitous language (each term: canonical definition + scope-type, original-language headword).\n- `pending\u002Fgaps.md` — punch-list of unresolved \u002F unverifiable items (claim · why · what's-needed · owner).\n- `.claude\u002Fskills\u002F\u003Cname>\u002FSKILL.md` — distilled, RED→GREEN-verified skill (symlink to `~\u002F.claude\u002Fskills\u002F` to make it live everywhere).\n- **Memory-store delta** (whenever a store exists) — every new\u002Frevised file (re)indexed with stale entries deleted first; every recalled\u002Ftouched\u002Fsuperseded entry reconciled; verified byproducts landed as point-in-time-verified entries.\n",{"name":5,"description":7},"main","c608f0990ee18432272458dd55b25b961c32501e",".claude\u002Fskills\u002Fdistill-domain-knowledge","https:\u002F\u002Fgithub.com\u002FYunxiangHuang\u002Fdistill-me","https:\u002F\u002Fgithub.com\u002FYunxiangHuang","https:\u002F\u002Fwuhenqs.com","2026-08-07 15:10:00",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add YunxiangHuang\u002Fdistill-me --skill distill-domain-knowledge",{"tool":37,"command":38},"bunx","bunx skills add YunxiangHuang\u002Fdistill-me --skill distill-domain-knowledge",{"tool":40,"command":41},"pnpm","pnpm dlx skills add YunxiangHuang\u002Fdistill-me --skill distill-domain-knowledge","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FYunxiangHuang\u002Fdistill-me\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdistill-domain-knowledge\u002FSKILL.md",true,[45,51,55,59,63],{"path":46,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":49,"sourceHtmlUrl":50},"references\u002Fglossary-first.md","REFERENCE","参考文档",3325,"https:\u002F\u002Fgithub.com\u002FYunxiangHuang\u002Fdistill-me\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdistill-domain-knowledge\u002Freferences\u002Fglossary-first.md",{"path":52,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":53,"sourceHtmlUrl":54},"references\u002Fgrilling.md",4768,"https:\u002F\u002Fgithub.com\u002FYunxiangHuang\u002Fdistill-me\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdistill-domain-knowledge\u002Freferences\u002Fgrilling.md",{"path":56,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":57,"sourceHtmlUrl":58},"references\u002Fmemory-loop.md",3184,"https:\u002F\u002Fgithub.com\u002FYunxiangHuang\u002Fdistill-me\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdistill-domain-knowledge\u002Freferences\u002Fmemory-loop.md",{"path":60,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":61,"sourceHtmlUrl":62},"references\u002Fred-green-validation.md",6059,"https:\u002F\u002Fgithub.com\u002FYunxiangHuang\u002Fdistill-me\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdistill-domain-knowledge\u002Freferences\u002Fred-green-validation.md",{"path":64,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":65,"sourceHtmlUrl":66},"references\u002Fverify-gate.md",4113,"https:\u002F\u002Fgithub.com\u002FYunxiangHuang\u002Fdistill-me\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fdistill-domain-knowledge\u002Freferences\u002Fverify-gate.md",[68,75,82,89],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":19},"01M77XD04SMV2Q2KYE3ET3DEJ3","recall-domainA-knowledge","Load distilled domainA knowledge before reasoning about domainA. Invoke BEFORE concluding.","https:\u002F\u002Fgithub.com\u002FYunxiangHuang\u002Fdistill-me\u002Fblob\u002Fmain\u002Ftest\u002Ffixtures\u002Fkb-domainA\u002F.claude\u002Fskills\u002Frecall-domainA-knowledge\u002FSKILL.md",[],[],{"skillNo":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M77XCVNA208MGP7G83GGN7XN","start-distill-domain-knowledge","Use to START distilling a NEW domain into a knowledge base — the per-domain 开张 wizard that scaffolds\u002Fextends the fixed KB structure, grills domain identity (boundary + core services), optionally scouts an initial backlog, then hands off to distill-domain-knowledge for per-slice work. Handles both greenfield (fresh KB) and additive (add a domain to an existing KB) via runtime grill.","https:\u002F\u002Fgithub.com\u002FYunxiangHuang\u002Fdistill-me\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fstart-distill-domain-knowledge\u002FSKILL.md",[],[],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":19},"01M77XCRD6T4XSMDBA6NXFP07A","grill-me","Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions \"grill me\".","https:\u002F\u002Fgithub.com\u002FYunxiangHuang\u002Fdistill-me\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fgrill-me\u002FSKILL.md",[],[],{"skillNo":90,"name":91,"nameZh":6,"description":92,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":93,"categories":94,"keywords":95,"publishedAt":19},"01M77XCHBBA3KQ8BMR1MMVSHM4","find-domain-knowledge","Discover and load distilled domain knowledge before reasoning about a business domain. Use when a sub-agent — especially one dispatched inside an orchestrated workflow (e.g. a multi-agent investigation \u002F RCA swarm), where the proactive skill-invocation discipline is skipped — is about to investigate, conclude about, or reason over a business domain (orders, payments, billing, lifecycle status, refunds, discounts, deposits, or any domain a project has distilled). Finds the matching `recall-*-knowledge` entry-point skill for whatever domain the current task touches and loads it, so you build on verified distilled mechanics instead of re-deriving them from raw source. Invoke this BEFORE concluding, not after.","https:\u002F\u002Fgithub.com\u002FYunxiangHuang\u002Fdistill-me\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Ffind-domain-knowledge\u002FSKILL.md",[],[]]