[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M797BPWENGAB1JRMA6Y38QH4":3,"skill-same:01M797BPWENGAB1JRMA6Y38QH4":55},{"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":25,"commitSha":26,"skillDirectory":27,"forksCount":28,"repositoryHtmlUrl":29,"repositoryHomepageUrl":30,"ownerHtmlUrl":31,"ownerBlogUrl":32,"ownerDescription":33,"updatedAt":34,"install":35,"resources":48},"01M797BPWENGAB1JRMA6Y38QH4","writing-rules",null,"Write or review anything an agent reads — a core rule, a profile gate, a SKILL.md description, an instruction-file line, a subagent prompt, a handoff note, a verify-phase label. Part of the Agentsmith harness; supplies the levers that decide whether a line changes behaviour or only costs tokens — the two loads, context pointers, the ladder, completion criteria, leading words, the no-op test. Its default move is deletion, so it earns most on a draft that already exists.","PromptPartner\u002Fagentsmith","PromptPartner","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F258588204?v=4","01M797A9ZHJQNDS3K4CCKMD99M","PromptPartner.AI","01M797AA04A34XHEQDF3XBX402",152,"MIT","https:\u002F\u002Fgithub.com\u002FPromptPartner\u002Fagentsmith\u002Fblob\u002Fmaster\u002Fskills\u002Fwriting-rules\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: writing-rules\ndescription: Write or review anything an agent reads — a core rule, a profile gate, a SKILL.md description, an instruction-file line, a subagent prompt, a handoff note, a verify-phase label. Part of the Agentsmith harness; supplies the levers that decide whether a line changes behaviour or only costs tokens — the two loads, context pointers, the ladder, completion criteria, leading words, the no-op test. Its default move is deletion, so it earns most on a draft that already exists.\ncompatibility: Requires an Agent Skills-compatible coding agent with filesystem access.\n---\n\n# Writing rules — and anything else an agent reads\n\nYou are writing for a reader who has already read everything. Explanation is waste; precision is\nthe whole job. Left alone, a model writing instructions for a model spends most of its words\nrestating what it already knows — every one of those lines is a **no-op**: paid every turn,\nchanging nothing.\n\nThe packaging differs (a `core\u002F` rule, a profile gate, a `SKILL.md`, a subagent prompt, a handoff\nnote) but the writing does not. The same levers make each one predictable, so the agent takes the\nsame *process* every run rather than producing the same output.\n\nWhen the document is a harness surface — a `core\u002F` rule, a profile, a skill, a hook — read\n[`HARNESS-SURFACES.md`](HARNESS-SURFACES.md) for where it goes and what it costs.\n\n## Instruction vocabulary\n`AGENTS.md` is the canonical portable instruction file. Never infer the active agent from this\nskill's install path: portable canonical content may be copied into a required client adapter such\nas `.claude\u002Fskills`. Treat client-specific instruction files as generated adapters; edit their\nshared `core\u002F` or `profiles\u002F` source, never an adapter copy.\n\n## The two loads\n\nEvery document and every pointer spends one of two budgets. Most authoring decisions are this one\ntrade made in different places:\n\n- **Context load** — what always-loaded material costs the agent's window: an instruction-file line, a\n  skill `description`, anything sitting in context every turn whether or not it fires.\n- **Cognitive load** — what it costs *you*: knowing which documents exist and when to reach for\n  each. You are the index. Not a cost to minimise — it's the price of human agency. Spend it where\n  human judgement matters; remove it where it doesn't.\n\nMaterial behind a pointer escapes context load for the price of the pointer's own line. Material\nwith no pointer at all rides entirely on cognitive load.\n\n## Context pointers\n\nA **context pointer** is a reference held in context that names out-of-context material and encodes\nwhen to reach it. A skill's `description` and an instruction-file line naming a doc are **the same\nobject**. The pointer's *wording*, not its target, decides how reliably the agent reaches through\nit — so a must-have target behind a weak pointer is a variance bug. Sharpen the wording first;\ninline the material only if sharpening fails.\n\nA pointer does two jobs: say what the material is, and list the **branches** that should trigger it\n(a branch is a distinct case the document handles). Every word is paid every turn, so prune it\nharder than the body:\n\n- **Front-load the leading word** — the pointer is where it does its triggering work.\n- **One trigger per branch.** Synonyms renaming one branch are one branch written twice.\n- **Cut identity the body already carries.**\n\n## The ladder — where each piece sits\n\nDocuments are built from **steps** (ordered actions) and **reference** (facts consulted on demand).\nThey mix freely: all steps, all reference, or both. The decision is how far down each piece sits:\n\n1. **In-file step** — what the agent does, in order. The primary tier.\n2. **In-file reference** — consulted on demand. Often a legitimately flat peer-set (every gate of a\n   profile on one rung). That's an arrangement, not a smell.\n3. **Disclosed reference** — a separate file behind a pointer, loaded only when the pointer fires.\n\n**Progressive disclosure** is the move down that ladder so the top stays legible. Branching is the\ncleanest test: **inline what every branch needs; disclose what only some branches reach.** Push too\nlittle down and the top bloats; push too much and you hide what the agent actually needs.\n\n**Co-location** is the within-file companion — the ladder decides how far down, co-location decides\nwhat sits beside it. Keep a concept's definition, rules and caveats under one heading so reading\none part brings its neighbours. (Distinct from duplication: that repeats one meaning in two places,\nscattering fragments one meaning across many.)\n\n**Sprawl** is the failure mode here — a document simply too long, even when every line is live and\nunique. Attention thins across the excess. The cure is the ladder, not tighter sentences.\n\n## Completion criteria\n\nEvery step ends on a **completion criterion** — the condition that says the work is done. Two\nproperties make it a lever, and both sharpen R5 (verify before you call it done):\n\n- **Clarity** — can the agent tell done from not-done? A vague bound (\"understanding reached\")\n  invites **premature completion**: ending early because attention slipped to *being done*. The\n  visible steps still ahead supply the pull; the criterion's clarity is the resistance. Sharpen the\n  bound first — it's local and cheap. Only if it's irreducibly fuzzy *and* you observe the rush,\n  split the sequence so the later steps aren't in view. That only works across a real context\n  boundary (a handoff, a subagent dispatch); an inline call clears nothing.\n- **Demand** — how much it requires. \"Every consumer of the change checked\" forces thorough work\n  where \"check the change\" does not. Demand drives **legwork**: the digging latent in the wording\n  rather than written as its own step. It isn't step-bound — \"every gate applied\" binds a body of\n  flat reference the same way.\n\nThe strongest criteria are both checkable and exhaustive. This is why R3 says *every* consumer and\nR2 says evidence *you produced* — both are demand, written into the bound.\n\n## Leading words\n\nA **leading word** is a compact concept already living in the model's pretraining that the agent\nthinks with while running the document (*atomic*, *evidence*, *Chesterton's Fence*, *tracer\nbullet*). Repeated as a token, never as a sentence, it accumulates a distributed definition and\nanchors a whole region of behaviour in the fewest tokens — because it recruits priors the model\nalready holds. Coining your own works if you define it clearly, but a made-up word recruits\nnothing: you pay in definition tokens what a pretrained word gives free. Reach for an existing word\nfirst.\n\nIt anchors twice — in the body for *execution* (the same behaviour every time the word appears),\nand in a pointer for *invocation* (when the same word lives in your prompts, your docs and your\ncode, the agent links them and reaches the material more reliably).\n\nHunt for passages begging to collapse into one token. A triad spelled out at three sites; a pointer\nspending a sentence to gesture at one idea:\n\n- \"fast, deterministic, low-overhead\" → *tight* (a *tight* loop).\n- \"a loop you believe in\" → *red* — a fuzzy gate becomes a binary observable state.\n\nAssume every document is carrying restatements that leading words retire. Go find them.\n\n**Negation** is the failure mode beside this lever. Steering by prohibition drags the forbidden\nbehaviour into context and makes it *more* available, not less — *don't think of an elephant*, and\nthe elephant is all there is. The negation is a weak modifier the strongly-activated concept\noverruns, so the ban half-reads as an instruction. **Prompt the positive**: state the target\nbehaviour so the banned one is never spoken. A prohibition earns its place only as a hard guardrail\nyou cannot phrase positively — and even then, pair it with the positive target.\n\n## Pruning — the default move is deletion\n\n- Keep each meaning in a **single source of truth**, so changing the behaviour is a one-place edit.\n  **Duplication** costs maintenance and tokens, and inflates a meaning's prominence past its real\n  rank. (The accidental inverse of a leading word, which repeats a token on purpose, never the\n  meaning.)\n- The **environment** is a source of truth too — `package.json` scripts, config files, `--help`\n  output, the directory layout. A document restating it is a **cache**, and a cache earns its load\n  only when the lookup is expensive. Cache what the agent cannot find by looking: the unwritten\n  convention, the reason behind a choice, the gotcha no config confesses. Leave one-command lookups\n  to the environment, where they cannot go stale.\n- Check every line for **relevance** — does it still bear on what the document does? A line loses\n  it by never bearing on the task, or by going stale as the world it describes changes. Without a\n  pruning discipline the default fate is **sediment**: stale layers that settle because adding\n  feels safe and removing feels risky.\n- Hunt **no-ops** sentence by sentence. The test is behavioural, not aesthetic: **delete the line —\n  does the agent behave differently?** If not, it was never a rule. The test is model-relative, so\n  two people disagreeing about a no-op disagree about the default, and settle it by *running the\n  document*, not by arguing. When a sentence fails, delete the whole sentence rather than trim\n  words from it. It grades leading words too: a word too weak to beat the default (*be thorough*,\n  when the agent is already thorough-ish) is itself a no-op, and the fix is a stronger word.\n\n## The review pass\n\nMost of the value lands here, not on a blank file. **Name the failure mode before you fix it** —\nsprawl, sediment, duplication, negation, premature completion, no-op — because the vocabulary is\nthe repair kit. Then, given a draft:\n\n1. **No-op sweep, sentence by sentence.** Grade each line by behaviour, not by length: an agent\n   told to shorten optimises for length, because length is what it can see.\n2. **Find the duplication.** Nothing stated twice, in any form. It's the most reliable sign a\n   document was never tested.\n3. **Check the ladder.** Anything only some branches reach goes behind a pointer.\n4. **Re-read the pointers** — one trigger per branch, leading word first. A branch that needs\n   material and has no pointer to it is the same finding in reverse.\n5. **State every prohibition as its positive target**, unless it's a hard guardrail.\n\n**Evidence when you cannot run it.** The no-op test is behavioural, so a review with no run behind\nit yields *candidates*, not verdicts (R2). The reachable offline bar: quote the line, name the\nfailure mode, and say which surviving line already carries the meaning. A deletion you can point at\na single source of truth for is evidence; one justified by \"reads redundant\" is a guess. Mark the\nguesses, and settle them by running the document.\n\nIt's working if the document got **shorter** as it got better; if a leading word is visibly doing\nwork in more than one place; and if nothing is stated twice.\n\n## Report\n\nSay what you deleted and why, in failure-mode terms — not just what you added. If the document grew,\njustify the growth against the two loads.\n\n---\n*Adapted from [`writing-for-agents`](https:\u002F\u002Fgithub.com\u002Fmattpocock\u002Fskills\u002Fblob\u002F5b15a47f2d7150f545fbcacbfe381787fc0230dc\u002Fskills\u002Fproductivity\u002Fwriting-for-agents\u002FSKILL.md)\nby Matt Pocock — MIT © 2026. Concepts and leading words are his; the harness framing and R-number\ncross-references are ours. See `docs\u002F18-influences.md`.*\n","# Writing rules — and anything else an agent reads\n\nYou are writing for a reader who has already read everything. Explanation is waste; precision is\nthe whole job. Left alone, a model writing instructions for a model spends most of its words\nrestating what it already knows — every one of those lines is a **no-op**: paid every turn,\nchanging nothing.\n\nThe packaging differs (a `core\u002F` rule, a profile gate, a `SKILL.md`, a subagent prompt, a handoff\nnote) but the writing does not. The same levers make each one predictable, so the agent takes the\nsame *process* every run rather than producing the same output.\n\nWhen the document is a harness surface — a `core\u002F` rule, a profile, a skill, a hook — read\n[`HARNESS-SURFACES.md`](HARNESS-SURFACES.md) for where it goes and what it costs.\n\n## Instruction vocabulary\n`AGENTS.md` is the canonical portable instruction file. Never infer the active agent from this\nskill's install path: portable canonical content may be copied into a required client adapter such\nas `.claude\u002Fskills`. Treat client-specific instruction files as generated adapters; edit their\nshared `core\u002F` or `profiles\u002F` source, never an adapter copy.\n\n## The two loads\n\nEvery document and every pointer spends one of two budgets. Most authoring decisions are this one\ntrade made in different places:\n\n- **Context load** — what always-loaded material costs the agent's window: an instruction-file line, a\n  skill `description`, anything sitting in context every turn whether or not it fires.\n- **Cognitive load** — what it costs *you*: knowing which documents exist and when to reach for\n  each. You are the index. Not a cost to minimise — it's the price of human agency. Spend it where\n  human judgement matters; remove it where it doesn't.\n\nMaterial behind a pointer escapes context load for the price of the pointer's own line. Material\nwith no pointer at all rides entirely on cognitive load.\n\n## Context pointers\n\nA **context pointer** is a reference held in context that names out-of-context material and encodes\nwhen to reach it. A skill's `description` and an instruction-file line naming a doc are **the same\nobject**. The pointer's *wording*, not its target, decides how reliably the agent reaches through\nit — so a must-have target behind a weak pointer is a variance bug. Sharpen the wording first;\ninline the material only if sharpening fails.\n\nA pointer does two jobs: say what the material is, and list the **branches** that should trigger it\n(a branch is a distinct case the document handles). Every word is paid every turn, so prune it\nharder than the body:\n\n- **Front-load the leading word** — the pointer is where it does its triggering work.\n- **One trigger per branch.** Synonyms renaming one branch are one branch written twice.\n- **Cut identity the body already carries.**\n\n## The ladder — where each piece sits\n\nDocuments are built from **steps** (ordered actions) and **reference** (facts consulted on demand).\nThey mix freely: all steps, all reference, or both. The decision is how far down each piece sits:\n\n1. **In-file step** — what the agent does, in order. The primary tier.\n2. **In-file reference** — consulted on demand. Often a legitimately flat peer-set (every gate of a\n   profile on one rung). That's an arrangement, not a smell.\n3. **Disclosed reference** — a separate file behind a pointer, loaded only when the pointer fires.\n\n**Progressive disclosure** is the move down that ladder so the top stays legible. Branching is the\ncleanest test: **inline what every branch needs; disclose what only some branches reach.** Push too\nlittle down and the top bloats; push too much and you hide what the agent actually needs.\n\n**Co-location** is the within-file companion — the ladder decides how far down, co-location decides\nwhat sits beside it. Keep a concept's definition, rules and caveats under one heading so reading\none part brings its neighbours. (Distinct from duplication: that repeats one meaning in two places,\nscattering fragments one meaning across many.)\n\n**Sprawl** is the failure mode here — a document simply too long, even when every line is live and\nunique. Attention thins across the excess. The cure is the ladder, not tighter sentences.\n\n## Completion criteria\n\nEvery step ends on a **completion criterion** — the condition that says the work is done. Two\nproperties make it a lever, and both sharpen R5 (verify before you call it done):\n\n- **Clarity** — can the agent tell done from not-done? A vague bound (\"understanding reached\")\n  invites **premature completion**: ending early because attention slipped to *being done*. The\n  visible steps still ahead supply the pull; the criterion's clarity is the resistance. Sharpen the\n  bound first — it's local and cheap. Only if it's irreducibly fuzzy *and* you observe the rush,\n  split the sequence so the later steps aren't in view. That only works across a real context\n  boundary (a handoff, a subagent dispatch); an inline call clears nothing.\n- **Demand** — how much it requires. \"Every consumer of the change checked\" forces thorough work\n  where \"check the change\" does not. Demand drives **legwork**: the digging latent in the wording\n  rather than written as its own step. It isn't step-bound — \"every gate applied\" binds a body of\n  flat reference the same way.\n\nThe strongest criteria are both checkable and exhaustive. This is why R3 says *every* consumer and\nR2 says evidence *you produced* — both are demand, written into the bound.\n\n## Leading words\n\nA **leading word** is a compact concept already living in the model's pretraining that the agent\nthinks with while running the document (*atomic*, *evidence*, *Chesterton's Fence*, *tracer\nbullet*). Repeated as a token, never as a sentence, it accumulates a distributed definition and\nanchors a whole region of behaviour in the fewest tokens — because it recruits priors the model\nalready holds. Coining your own works if you define it clearly, but a made-up word recruits\nnothing: you pay in definition tokens what a pretrained word gives free. Reach for an existing word\nfirst.\n\nIt anchors twice — in the body for *execution* (the same behaviour every time the word appears),\nand in a pointer for *invocation* (when the same word lives in your prompts, your docs and your\ncode, the agent links them and reaches the material more reliably).\n\nHunt for passages begging to collapse into one token. A triad spelled out at three sites; a pointer\nspending a sentence to gesture at one idea:\n\n- \"fast, deterministic, low-overhead\" → *tight* (a *tight* loop).\n- \"a loop you believe in\" → *red* — a fuzzy gate becomes a binary observable state.\n\nAssume every document is carrying restatements that leading words retire. Go find them.\n\n**Negation** is the failure mode beside this lever. Steering by prohibition drags the forbidden\nbehaviour into context and makes it *more* available, not less — *don't think of an elephant*, and\nthe elephant is all there is. The negation is a weak modifier the strongly-activated concept\noverruns, so the ban half-reads as an instruction. **Prompt the positive**: state the target\nbehaviour so the banned one is never spoken. A prohibition earns its place only as a hard guardrail\nyou cannot phrase positively — and even then, pair it with the positive target.\n\n## Pruning — the default move is deletion\n\n- Keep each meaning in a **single source of truth**, so changing the behaviour is a one-place edit.\n  **Duplication** costs maintenance and tokens, and inflates a meaning's prominence past its real\n  rank. (The accidental inverse of a leading word, which repeats a token on purpose, never the\n  meaning.)\n- The **environment** is a source of truth too — `package.json` scripts, config files, `--help`\n  output, the directory layout. A document restating it is a **cache**, and a cache earns its load\n  only when the lookup is expensive. Cache what the agent cannot find by looking: the unwritten\n  convention, the reason behind a choice, the gotcha no config confesses. Leave one-command lookups\n  to the environment, where they cannot go stale.\n- Check every line for **relevance** — does it still bear on what the document does? A line loses\n  it by never bearing on the task, or by going stale as the world it describes changes. Without a\n  pruning discipline the default fate is **sediment**: stale layers that settle because adding\n  feels safe and removing feels risky.\n- Hunt **no-ops** sentence by sentence. The test is behavioural, not aesthetic: **delete the line —\n  does the agent behave differently?** If not, it was never a rule. The test is model-relative, so\n  two people disagreeing about a no-op disagree about the default, and settle it by *running the\n  document*, not by arguing. When a sentence fails, delete the whole sentence rather than trim\n  words from it. It grades leading words too: a word too weak to beat the default (*be thorough*,\n  when the agent is already thorough-ish) is itself a no-op, and the fix is a stronger word.\n\n## The review pass\n\nMost of the value lands here, not on a blank file. **Name the failure mode before you fix it** —\nsprawl, sediment, duplication, negation, premature completion, no-op — because the vocabulary is\nthe repair kit. Then, given a draft:\n\n1. **No-op sweep, sentence by sentence.** Grade each line by behaviour, not by length: an agent\n   told to shorten optimises for length, because length is what it can see.\n2. **Find the duplication.** Nothing stated twice, in any form. It's the most reliable sign a\n   document was never tested.\n3. **Check the ladder.** Anything only some branches reach goes behind a pointer.\n4. **Re-read the pointers** — one trigger per branch, leading word first. A branch that needs\n   material and has no pointer to it is the same finding in reverse.\n5. **State every prohibition as its positive target**, unless it's a hard guardrail.\n\n**Evidence when you cannot run it.** The no-op test is behavioural, so a review with no run behind\nit yields *candidates*, not verdicts (R2). The reachable offline bar: quote the line, name the\nfailure mode, and say which surviving line already carries the meaning. A deletion you can point at\na single source of truth for is evidence; one justified by \"reads redundant\" is a guess. Mark the\nguesses, and settle them by running the document.\n\nIt's working if the document got **shorter** as it got better; if a leading word is visibly doing\nwork in more than one place; and if nothing is stated twice.\n\n## Report\n\nSay what you deleted and why, in failure-mode terms — not just what you added. If the document grew,\njustify the growth against the two loads.\n\n---\n*Adapted from [`writing-for-agents`](https:\u002F\u002Fgithub.com\u002Fmattpocock\u002Fskills\u002Fblob\u002F5b15a47f2d7150f545fbcacbfe381787fc0230dc\u002Fskills\u002Fproductivity\u002Fwriting-for-agents\u002FSKILL.md)\nby Matt Pocock — MIT © 2026. Concepts and leading words are his; the harness framing and R-number\ncross-references are ours. See `docs\u002F18-influences.md`.*\n",{"name":5,"description":7,"compatibility":24},"Requires an Agent Skills-compatible coding agent with filesystem access.","master","610914b5c8c24904de0259d70b74cb0d9b0289a4","skills\u002Fwriting-rules",7,"https:\u002F\u002Fgithub.com\u002FPromptPartner\u002Fagentsmith","https:\u002F\u002Fagentsmith.diy","https:\u002F\u002Fgithub.com\u002FPromptPartner","https:\u002F\u002Fpromptpartner.ai","We build enterprise-grade AI Agent systems, mostly for companies with 50M++ ARR. And you own the systems.","2026-11-15 00:04:54",{"source":8,"skillName":5,"commands":36,"prompt":46,"zipAvailable":47},[37,40,43],{"tool":38,"command":39},"npx","npx skills add PromptPartner\u002Fagentsmith --skill writing-rules",{"tool":41,"command":42},"bunx","bunx skills add PromptPartner\u002Fagentsmith --skill writing-rules",{"tool":44,"command":45},"pnpm","pnpm dlx skills add PromptPartner\u002Fagentsmith --skill writing-rules","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FPromptPartner\u002Fagentsmith\u002Fblob\u002Fmaster\u002Fskills\u002Fwriting-rules\u002FSKILL.md",true,[49],{"path":50,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":53,"sourceHtmlUrl":54},"HARNESS-SURFACES.md","REFERENCE","参考文档",6399,"https:\u002F\u002Fgithub.com\u002FPromptPartner\u002Fagentsmith\u002Fblob\u002Fmaster\u002Fskills\u002Fwriting-rules\u002FHARNESS-SURFACES.md",[56,63,68,75,82,89],{"skillNo":57,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":19},"01M797BYCC18F1KX1DT4ZNQC09","portable-fixture","Fixture proving that skill identity comes from metadata rather than its installation path.","https:\u002F\u002Fgithub.com\u002FPromptPartner\u002Fagentsmith\u002Fblob\u002Fmaster\u002Ftests\u002Ffixtures\u002Fcompatibility\u002Fskill-paths\u002Fclaude-adapter\u002F.claude\u002Fskills\u002Fportable-fixture\u002FSKILL.md",[],[],{"skillNo":64,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":67,"publishedAt":19},"01M797BVHW9SBC34MX3Y069GH5","https:\u002F\u002Fgithub.com\u002FPromptPartner\u002Fagentsmith\u002Fblob\u002Fmaster\u002Ftests\u002Ffixtures\u002Fcompatibility\u002Fskill-paths\u002Fcanonical\u002F.agents\u002Fskills\u002Fportable-fixture\u002FSKILL.md",[],[],{"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},"01M797BKXS4QT9HQDF119CRQCZ","wayfinder","Wayfind a foggy, multi-session effort into an accepted repository spec and separate implementation-ticket drafts. Use when the destination is known but important decisions, dependencies, or scope are not yet clear enough to implement safely.","https:\u002F\u002Fgithub.com\u002FPromptPartner\u002Fagentsmith\u002Fblob\u002Fmaster\u002Fskills\u002Fwayfinder\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},"01M797BGEMVXWZ4RHSJRTNJ71W","verify","Answer \"is this shippable \u002F done?\" with evidence before a commit or PR — fires on \"verify\", \"is this done\", \"ready to ship?\", pre-commit or pre-PR. Part of the Agentsmith harness; runs the project's verify phases and never claims \"passing\" without showing the output (R5).","https:\u002F\u002Fgithub.com\u002FPromptPartner\u002Fagentsmith\u002Fblob\u002Fmaster\u002Fskills\u002Fverify\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},"01M797BCSTDN8M8TM2HTBWRC60","new-research","Start a durable research \u002F source note that captures what you read and found — fires on \"start a research note\", \"capture these sources\", \"log my research\". Part of the Agentsmith harness; scaffolds docs\u002Fresearch\u002F\u003Cslug>.md so findings live in the repo, never in disposable memory (R9), and are never silently deleted.","https:\u002F\u002Fgithub.com\u002FPromptPartner\u002Fagentsmith\u002Fblob\u002Fmaster\u002Fskills\u002Fnew-research\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},"01M797B8RSE0FF78DX50H2VPC6","new-feedback","Log a harness lesson \u002F post-incident so the SYSTEM improves, not just this one bug — fires on \"log a harness lesson\", \"post-incident\", \"we should make this less likely\", System-Evolution moments. Part of the Agentsmith harness; scaffolds a numbered docs\u002Ffeedback\u002FNNNN-*.md with the five-stage template (R9 — numbers are never reused).","https:\u002F\u002Fgithub.com\u002FPromptPartner\u002Fagentsmith\u002Fblob\u002Fmaster\u002Fskills\u002Fnew-feedback\u002FSKILL.md",[],[]]