[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M78D66R7FSV4E613YS25N16F":3,"skill-same:01M78D66R7FSV4E613YS25N16F":57},{"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":6,"ownerHtmlUrl":30,"ownerBlogUrl":6,"ownerDescription":31,"updatedAt":32,"install":33,"resources":46},"01M78D66R7FSV4E613YS25N16F","codebase-design",null,"Before writing a new interface (module boundary, class API, function signature, or service contract) that at least two call sites will depend on, that crosses a module\u002Fservice\u002Fteam\u002Fpersistence boundary, or that will be expensive to change later — produce 3+ radically different candidate designs and compare them on depth, locality, and seam placement before picking one. Use on \"before committing to an interface\", \"design this API\u002Fmodule\u002Fclass boundary\", \"how should this be structured\", \"compare interface designs\", \"is this the right abstraction\", \"design it twice\", reviewing a proposed interface shape in a PR — or self-trigger whenever about to write a new interface meeting that bar.","JRichlen\u002Fagent-plugins","JRichlen","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F9574264?v=4","01M78D3QWDTGRFSC2A9XADS6GM","Jordan Richlen","01M78D3QX3W2ZQEEACQHKDX3HW",1,"MIT","https:\u002F\u002Fgithub.com\u002FJRichlen\u002Fagent-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcodebase-design\u002Fskills\u002Fcodebase-design\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: codebase-design\ndescription: >-\n  Before writing a new interface (module boundary, class API, function\n  signature, or service contract) that at least two call sites will depend\n  on, that crosses a module\u002Fservice\u002Fteam\u002Fpersistence boundary, or that will\n  be expensive to change later — produce 3+ radically different candidate\n  designs and compare them on depth, locality, and seam placement before\n  picking one. Use on \"before committing to an interface\", \"design this\n  API\u002Fmodule\u002Fclass boundary\", \"how should this be structured\", \"compare\n  interface designs\", \"is this the right abstraction\", \"design it twice\",\n  reviewing a proposed interface shape in a PR — or self-trigger whenever\n  about to write a new interface meeting that bar.\nlicense: MIT\ncompatibility: >-\n  PORTABILITY: harness-agnostic. The core design-it-twice procedure is plain\n  reasoning and prose comparison — no subagent-spawning tool, no Workflow\n  tool, no hooks, no harness-specific primitive required. An optional\n  parallel-subagent escalation exists only where a subagent-spawning tool is\n  present (see the Portability section below) — a convenience, not a\n  dependency, mirroring voice's \"hook is a convenience, not a dependency\"\n  pattern rather than second-opinion's hard gate on subagents.\n---\n\n# codebase-design\n\n## Invariant\n\nBefore committing to an interface (a module boundary, class API, or function signature that ≥2 call sites will depend on, or that crosses a service\u002Fteam\u002Fpersistence boundary, or that is expensive to change later): ALWAYS produce 3+ radically different candidate designs and compare them on depth, locality, and seam placement before picking one — NEVER let the first workable interface ship unexamined. The chosen interface must ALWAYS hide its implementation complexity behind a boundary that tests hold honest at a confirmed seam — NEVER a shallow pass-through whose interface exists only to make internals swappable, and NEVER a test written against an unconfirmed seam.\n\n## Not this\n\nThree plugins in this marketplace share enough vocabulary to be reached for\nby mistake. None of them is this skill.\n\n- **wave-research** (`plugins\u002Fwave-research\u002Fskills\u002Fwave-research`) ships a\n  staged research method and two Workflow-tool templates for fanning subagents\n  out over RESEARCH dimensions and adversarially verifying the CLAIMS that\n  research surfaces — frozen ground-truth context, per-stage JSON schemas, a\n  verifier that defaults to disbelief. Hand-rolling subagents to research\n  several options and fact-check what they report back is its job.\n  codebase-design borrows only the *shape* of \"produce several, then\n  compare\", applied instead to DESIGN ALTERNATIVES the agent itself generates\n  and scores against fixed axes, never to claims a fan-out surfaced — no\n  Workflow tool, per-stage schema, or adversarial verifier required by\n  default (Step 1's optional parallel dispatch borrows the fan-out habit, not\n  the verify\u002Fharvest\u002Freconcile machinery). Reach for wave-research to research\n  N options and verify what's claimed about them; reach for this skill when\n  the shape space is already known and the job is to generate and score\n  candidates directly.\n- **second-opinion** (`plugins\u002Fvoice\u002Fskills\u002Fsecond-opinion`) is *post-hoc and\n  offer-only*: it validates a verdict that already exists and is forbidden\n  from emitting output unless subagents actually dispatched and reported\n  back. codebase-design is *pre-hoc and self-triggering* on the\n  interface-shaped heuristic in Step 0 — it runs BEFORE an interface is\n  written, produces designs rather than validating one that already exists,\n  and never needs a subagent-spawning tool to do its core job.\n- **grill-me** (`plugins\u002Fgrill-me\u002Fskills\u002Fgrill-me`) is the closest surface\n  match — it too runs \"before starting a nontrivial multi-step change whose\n  design isn't yet settled.\" But grill-me is a live, single-session\n  CONVERSATIONAL INTERVIEW of the USER aimed at reaching shared understanding\n  of a PLAN. codebase-design never interviews the user as its core mechanism\n  — it is the agent GENERATING and self-comparing 3+ concrete interface\n  designs against objective axes, with no requirement that the user\n  participate turn-by-turn. The two are complementary: grill-me's frontier\n  loop can reach a branch that IS an interface decision and hand off to\n  design-it-twice there instead of asking the user to freehand an API in\n  conversation — but the two must never merge into one \"design conversation\"\n  skill that does neither job well.\n\n## Step 0 — Is this interface-shaped?\n\nApply this heuristic before doing anything else — it's the gate that stops\noverhead on trivial code. The decision is interface-shaped, and the rest of\nthis procedure applies, if ANY of these hold:\n\n- **≥2 call sites** will depend on this shape (not counting the one you're\n  writing right now).\n- It **crosses a boundary** — a module, service, team, or persistence\n  boundary: a public API, a network contract, a DB schema, a package export.\n- It would be **expensive or breaking to change later** (this is\n  semver-gate's MAJOR-class test in miniature: does changing this later\n  break a contract someone else depends on?).\n- You are about to write a **genuinely new abstraction** — roughly 30+ lines\n  of new module\u002Fclass scaffolding — not extending an existing,\n  already-judged one.\n\nIf NONE hold — a private helper, a single call site, an internal\nimplementation detail nobody else touches — say so in one line (\"trivial,\nnot interface-shaped, skipping design-it-twice\") and go straight to\nimplementation. Do not manufacture a 3-design comparison for a getter. This\nmirrors grill-me's \"if the whole plan triages LIGHT end to end, say so\nbriefly and move on\" discipline — the same anti-overhead norm, aimed at\ninterface decisions instead of whole plans.\n\n## Step 1 — Produce 3+ radically different designs\n\nLoad `references\u002Fdesign-it-twice.md` now. Generate at least 3 candidate\ninterfaces for the SAME piece of functionality. \"Radically different\" is a\ncheckable bar, not a vibe: each candidate must differ from every other\ncandidate on at least one of the three axes in Step 2 by more than a rename\nor parameter reorder. Four generator prompts force real divergence — use at\nleast 2 of them across the 3 candidates so they don't converge on one shape:\n\n1. Vary **where state lives** — fully inside the module, passed in by the\n   caller each call, or externalized to a shared store.\n2. Vary the **call shape** — a single synchronous call, event\u002Fmessage-driven,\n   or declarative\u002Fconfig-driven (the caller states the desired end-state,\n   the module figures out how).\n3. Vary the **surface size** — one fat interface exposing every step, one\n   thin single-entry facade, or a composable pipeline of small pieces.\n4. Vary **who owns error\u002Fretry\u002Fedge-case handling** — the caller, the\n   callee, or a shared policy object both consult.\n\nThe default path needs no subagent-spawning tool: generate all 3 designs\nyourself, sequentially, in this session, explicitly trying to make each one\ndefensible on its own terms before moving to the next (don't let candidate 2\nbe a strawman built to lose to candidate 1). If a subagent-spawning tool\nhappens to be available, you MAY optionally dispatch each candidate to a\nseparate agent in parallel for genuine independence — that's an optimization\nborrowed from wave-research's fan-out habit (see \"Not this\"), never a\ndependency; see Portability below.\n\n## Step 2 — Compare on depth, locality, seam placement\n\nStill from `references\u002Fdesign-it-twice.md` — full scoring questions and\nworked examples live there. For EACH candidate, answer these three named\nquestions — the \"exact check\" that replaces \"be careful\":\n\n1. **DEPTH** — ratio of caller-visible footprint to hidden decision count. A\n   small footprint hiding a lot is deep (good); a footprint roughly 1:1 with\n   what it does is shallow.\n2. **LOCALITY** — how many call sites\u002Ffiles a plausible bug fix or future\n   change would touch, and whether the caller must hold a cross-call\n   invariant in their head. Fewer forced call sites and no caller-held\n   invariants is better.\n3. **SEAM PLACEMENT** — apply the deletion test (full definition in\n   `references\u002Fdeep-modules.md`): mentally delete the module and inline its\n   logic at each call site. Real complexity disappearing from the codebase is\n   a real seam; code that reads almost identically minus one layer of\n   indirection is a pass-through, not a seam.\n\nBuild a 3-candidates x 3-axes comparison — a short table or three short\nparagraphs, whichever the answer's complexity warrants — then pick a winner\nand write ONE paragraph of rationale that cites the specific depth\u002Flocality\u002F\nseam answers that decided it. Never \"this one felt cleaner.\" If two\ncandidates tie, **locality wins the tiebreak**: a design that concentrates\nchange in fewer places compounds in your favor over the module's lifetime\nmore than a marginal depth or seam difference does.\n\n## Step 3 — Confirm seams before writing any implementation\n\nLoad `references\u002Fdeep-modules.md` now (Seam-Based Design & Test Agreement\nsubsection). For the CHOSEN design only, name each seam explicitly as a\nshort bullet list — \"this call boundary is a confirmed seam; tests double\nhere and nowhere else.\" Apply the two-adapter check from the same reference\nto each named seam: can you name a second concrete implementation this\nboundary would have to support (a fake for tests, a different backend, a v2\nprotocol)? If yes, the seam is real — keep it. If you can only imagine one\nimplementation ever existing, it's a hypothetical seam: either find the real\nexternal boundary further out and move the seam there, or drop the boundary\nand let the code be direct. A test written against a boundary that fails\nthis check does not earn its place — per the Seam-Based Design & Test\nAgreement, no test at an unconfirmed seam — flag it as a smell rather than\nwriting it.\n\n## Step 4 — Final depth check before handoff\n\nStill in `references\u002Fdeep-modules.md` (Module Depth Analysis subsection) —\nfull category definitions live there. Classify the chosen design's\ndependencies into four categories to decide what gets a test double and what\ndoesn't:\n\n1. **Pure\u002Finternal logic** — no I\u002FO, deterministic given inputs. Test\n   directly, no mocking.\n2. **Owned deep dependency** — another module in this codebase with its own\n   confirmed seam and tests. Call it for real, don't re-mock it here.\n3. **Unowned external boundary** — a third-party API, the network, the\n   filesystem, the clock, another team's service. This is where a test\n   double belongs, and the only place one belongs.\n4. **Config\u002Fenvironment surface** — values that vary by deploy, not by\n   logic. Inject as parameters; never mock a getter.\n\nDiagnostic — the over-fragmentation check: if the chosen interface exists\nONLY so a test can swap out a category-1 or category-2 dependency, that's\nexactly the thin, testability-only interface failure mode. Collapse it back\ninto its caller and push the test out to the real category-3 boundary\ninstead. A module whose only adapter will ever be itself just failed the\ntwo-adapter check from Step 3 as well — the two diagnostics should agree.\n\n## Step 5 — Ship the decision, not the debate\n\nOutput is: the chosen design (interface signature\u002Fshape), the one-paragraph\nrationale from Step 2, the named-seams list from Step 3, and nothing about\nthe 2+ rejected designs beyond a one-line \"considered and rejected: X\n(shallow — Y), Z (poor locality — W)\" — enough for someone reading the\ndecision later to know it wasn't the first idea, without carrying the full\ncomparison forward. Implementation itself is out of scope for this skill —\nit hands off to whatever writes the code\u002Ftests next. Design-then-implement\nis a front-loaded, distinct step, not interleaved with coding.\n\n## Portability\n\nDefault mode needs NO subagent-spawning tool (see Step 1) — the\nalways-available path, matching grill-me's \"no subagents required\"\nportability class, not second-opinion's hard gate. Full detail on the\noptional parallel-dispatch escalation, where a subagent-spawning tool is\navailable, lives in `references\u002Fdesign-it-twice.md` (\"Optional: parallel\ngeneration\"). The escalation is always optional and never changes Steps 2-5.\n","# codebase-design\n\n## Invariant\n\nBefore committing to an interface (a module boundary, class API, or function signature that ≥2 call sites will depend on, or that crosses a service\u002Fteam\u002Fpersistence boundary, or that is expensive to change later): ALWAYS produce 3+ radically different candidate designs and compare them on depth, locality, and seam placement before picking one — NEVER let the first workable interface ship unexamined. The chosen interface must ALWAYS hide its implementation complexity behind a boundary that tests hold honest at a confirmed seam — NEVER a shallow pass-through whose interface exists only to make internals swappable, and NEVER a test written against an unconfirmed seam.\n\n## Not this\n\nThree plugins in this marketplace share enough vocabulary to be reached for\nby mistake. None of them is this skill.\n\n- **wave-research** (`plugins\u002Fwave-research\u002Fskills\u002Fwave-research`) ships a\n  staged research method and two Workflow-tool templates for fanning subagents\n  out over RESEARCH dimensions and adversarially verifying the CLAIMS that\n  research surfaces — frozen ground-truth context, per-stage JSON schemas, a\n  verifier that defaults to disbelief. Hand-rolling subagents to research\n  several options and fact-check what they report back is its job.\n  codebase-design borrows only the *shape* of \"produce several, then\n  compare\", applied instead to DESIGN ALTERNATIVES the agent itself generates\n  and scores against fixed axes, never to claims a fan-out surfaced — no\n  Workflow tool, per-stage schema, or adversarial verifier required by\n  default (Step 1's optional parallel dispatch borrows the fan-out habit, not\n  the verify\u002Fharvest\u002Freconcile machinery). Reach for wave-research to research\n  N options and verify what's claimed about them; reach for this skill when\n  the shape space is already known and the job is to generate and score\n  candidates directly.\n- **second-opinion** (`plugins\u002Fvoice\u002Fskills\u002Fsecond-opinion`) is *post-hoc and\n  offer-only*: it validates a verdict that already exists and is forbidden\n  from emitting output unless subagents actually dispatched and reported\n  back. codebase-design is *pre-hoc and self-triggering* on the\n  interface-shaped heuristic in Step 0 — it runs BEFORE an interface is\n  written, produces designs rather than validating one that already exists,\n  and never needs a subagent-spawning tool to do its core job.\n- **grill-me** (`plugins\u002Fgrill-me\u002Fskills\u002Fgrill-me`) is the closest surface\n  match — it too runs \"before starting a nontrivial multi-step change whose\n  design isn't yet settled.\" But grill-me is a live, single-session\n  CONVERSATIONAL INTERVIEW of the USER aimed at reaching shared understanding\n  of a PLAN. codebase-design never interviews the user as its core mechanism\n  — it is the agent GENERATING and self-comparing 3+ concrete interface\n  designs against objective axes, with no requirement that the user\n  participate turn-by-turn. The two are complementary: grill-me's frontier\n  loop can reach a branch that IS an interface decision and hand off to\n  design-it-twice there instead of asking the user to freehand an API in\n  conversation — but the two must never merge into one \"design conversation\"\n  skill that does neither job well.\n\n## Step 0 — Is this interface-shaped?\n\nApply this heuristic before doing anything else — it's the gate that stops\noverhead on trivial code. The decision is interface-shaped, and the rest of\nthis procedure applies, if ANY of these hold:\n\n- **≥2 call sites** will depend on this shape (not counting the one you're\n  writing right now).\n- It **crosses a boundary** — a module, service, team, or persistence\n  boundary: a public API, a network contract, a DB schema, a package export.\n- It would be **expensive or breaking to change later** (this is\n  semver-gate's MAJOR-class test in miniature: does changing this later\n  break a contract someone else depends on?).\n- You are about to write a **genuinely new abstraction** — roughly 30+ lines\n  of new module\u002Fclass scaffolding — not extending an existing,\n  already-judged one.\n\nIf NONE hold — a private helper, a single call site, an internal\nimplementation detail nobody else touches — say so in one line (\"trivial,\nnot interface-shaped, skipping design-it-twice\") and go straight to\nimplementation. Do not manufacture a 3-design comparison for a getter. This\nmirrors grill-me's \"if the whole plan triages LIGHT end to end, say so\nbriefly and move on\" discipline — the same anti-overhead norm, aimed at\ninterface decisions instead of whole plans.\n\n## Step 1 — Produce 3+ radically different designs\n\nLoad `references\u002Fdesign-it-twice.md` now. Generate at least 3 candidate\ninterfaces for the SAME piece of functionality. \"Radically different\" is a\ncheckable bar, not a vibe: each candidate must differ from every other\ncandidate on at least one of the three axes in Step 2 by more than a rename\nor parameter reorder. Four generator prompts force real divergence — use at\nleast 2 of them across the 3 candidates so they don't converge on one shape:\n\n1. Vary **where state lives** — fully inside the module, passed in by the\n   caller each call, or externalized to a shared store.\n2. Vary the **call shape** — a single synchronous call, event\u002Fmessage-driven,\n   or declarative\u002Fconfig-driven (the caller states the desired end-state,\n   the module figures out how).\n3. Vary the **surface size** — one fat interface exposing every step, one\n   thin single-entry facade, or a composable pipeline of small pieces.\n4. Vary **who owns error\u002Fretry\u002Fedge-case handling** — the caller, the\n   callee, or a shared policy object both consult.\n\nThe default path needs no subagent-spawning tool: generate all 3 designs\nyourself, sequentially, in this session, explicitly trying to make each one\ndefensible on its own terms before moving to the next (don't let candidate 2\nbe a strawman built to lose to candidate 1). If a subagent-spawning tool\nhappens to be available, you MAY optionally dispatch each candidate to a\nseparate agent in parallel for genuine independence — that's an optimization\nborrowed from wave-research's fan-out habit (see \"Not this\"), never a\ndependency; see Portability below.\n\n## Step 2 — Compare on depth, locality, seam placement\n\nStill from `references\u002Fdesign-it-twice.md` — full scoring questions and\nworked examples live there. For EACH candidate, answer these three named\nquestions — the \"exact check\" that replaces \"be careful\":\n\n1. **DEPTH** — ratio of caller-visible footprint to hidden decision count. A\n   small footprint hiding a lot is deep (good); a footprint roughly 1:1 with\n   what it does is shallow.\n2. **LOCALITY** — how many call sites\u002Ffiles a plausible bug fix or future\n   change would touch, and whether the caller must hold a cross-call\n   invariant in their head. Fewer forced call sites and no caller-held\n   invariants is better.\n3. **SEAM PLACEMENT** — apply the deletion test (full definition in\n   `references\u002Fdeep-modules.md`): mentally delete the module and inline its\n   logic at each call site. Real complexity disappearing from the codebase is\n   a real seam; code that reads almost identically minus one layer of\n   indirection is a pass-through, not a seam.\n\nBuild a 3-candidates x 3-axes comparison — a short table or three short\nparagraphs, whichever the answer's complexity warrants — then pick a winner\nand write ONE paragraph of rationale that cites the specific depth\u002Flocality\u002F\nseam answers that decided it. Never \"this one felt cleaner.\" If two\ncandidates tie, **locality wins the tiebreak**: a design that concentrates\nchange in fewer places compounds in your favor over the module's lifetime\nmore than a marginal depth or seam difference does.\n\n## Step 3 — Confirm seams before writing any implementation\n\nLoad `references\u002Fdeep-modules.md` now (Seam-Based Design & Test Agreement\nsubsection). For the CHOSEN design only, name each seam explicitly as a\nshort bullet list — \"this call boundary is a confirmed seam; tests double\nhere and nowhere else.\" Apply the two-adapter check from the same reference\nto each named seam: can you name a second concrete implementation this\nboundary would have to support (a fake for tests, a different backend, a v2\nprotocol)? If yes, the seam is real — keep it. If you can only imagine one\nimplementation ever existing, it's a hypothetical seam: either find the real\nexternal boundary further out and move the seam there, or drop the boundary\nand let the code be direct. A test written against a boundary that fails\nthis check does not earn its place — per the Seam-Based Design & Test\nAgreement, no test at an unconfirmed seam — flag it as a smell rather than\nwriting it.\n\n## Step 4 — Final depth check before handoff\n\nStill in `references\u002Fdeep-modules.md` (Module Depth Analysis subsection) —\nfull category definitions live there. Classify the chosen design's\ndependencies into four categories to decide what gets a test double and what\ndoesn't:\n\n1. **Pure\u002Finternal logic** — no I\u002FO, deterministic given inputs. Test\n   directly, no mocking.\n2. **Owned deep dependency** — another module in this codebase with its own\n   confirmed seam and tests. Call it for real, don't re-mock it here.\n3. **Unowned external boundary** — a third-party API, the network, the\n   filesystem, the clock, another team's service. This is where a test\n   double belongs, and the only place one belongs.\n4. **Config\u002Fenvironment surface** — values that vary by deploy, not by\n   logic. Inject as parameters; never mock a getter.\n\nDiagnostic — the over-fragmentation check: if the chosen interface exists\nONLY so a test can swap out a category-1 or category-2 dependency, that's\nexactly the thin, testability-only interface failure mode. Collapse it back\ninto its caller and push the test out to the real category-3 boundary\ninstead. A module whose only adapter will ever be itself just failed the\ntwo-adapter check from Step 3 as well — the two diagnostics should agree.\n\n## Step 5 — Ship the decision, not the debate\n\nOutput is: the chosen design (interface signature\u002Fshape), the one-paragraph\nrationale from Step 2, the named-seams list from Step 3, and nothing about\nthe 2+ rejected designs beyond a one-line \"considered and rejected: X\n(shallow — Y), Z (poor locality — W)\" — enough for someone reading the\ndecision later to know it wasn't the first idea, without carrying the full\ncomparison forward. Implementation itself is out of scope for this skill —\nit hands off to whatever writes the code\u002Ftests next. Design-then-implement\nis a front-loaded, distinct step, not interleaved with coding.\n\n## Portability\n\nDefault mode needs NO subagent-spawning tool (see Step 1) — the\nalways-available path, matching grill-me's \"no subagents required\"\nportability class, not second-opinion's hard gate. Full detail on the\noptional parallel-dispatch escalation, where a subagent-spawning tool is\navailable, lives in `references\u002Fdesign-it-twice.md` (\"Optional: parallel\ngeneration\"). The escalation is always optional and never changes Steps 2-5.\n",{"name":5,"license":15,"description":7,"compatibility":24},"PORTABILITY: harness-agnostic. The core design-it-twice procedure is plain reasoning and prose comparison — no subagent-spawning tool, no Workflow tool, no hooks, no harness-specific primitive required. An optional parallel-subagent escalation exists only where a subagent-spawning tool is present (see the Portability section below) — a convenience, not a dependency, mirroring voice's \"hook is a convenience, not a dependency\" pattern rather than second-opinion's hard gate on subagents.","main","929887017aa19a575de99be4e3f9b78d0d7ad48d","plugins\u002Fcodebase-design\u002Fskills\u002Fcodebase-design",0,"https:\u002F\u002Fgithub.com\u002FJRichlen\u002Fagent-plugins","https:\u002F\u002Fgithub.com\u002FJRichlen","Solutions Architect","2026-11-12 14:00:54",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add JRichlen\u002Fagent-plugins --skill codebase-design",{"tool":39,"command":40},"bunx","bunx skills add JRichlen\u002Fagent-plugins --skill codebase-design",{"tool":42,"command":43},"pnpm","pnpm dlx skills add JRichlen\u002Fagent-plugins --skill codebase-design","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FJRichlen\u002Fagent-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcodebase-design\u002Fskills\u002Fcodebase-design\u002FSKILL.md",true,[47,53],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"references\u002Fdeep-modules.md","REFERENCE","参考文档",7509,"https:\u002F\u002Fgithub.com\u002FJRichlen\u002Fagent-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcodebase-design\u002Fskills\u002Fcodebase-design\u002Freferences\u002Fdeep-modules.md",{"path":54,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Fdesign-it-twice.md",10993,"https:\u002F\u002Fgithub.com\u002FJRichlen\u002Fagent-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fcodebase-design\u002Fskills\u002Fcodebase-design\u002Freferences\u002Fdesign-it-twice.md",[58,65,72,79,86,93],{"skillNo":59,"name":60,"nameZh":6,"description":61,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":62,"categories":63,"keywords":64,"publishedAt":19},"01M78DCJ7N0S5GVNPR0849HPRA","wayfinder","Chart a multi-session effort as a labeled map of typed decision tickets (grilling \u002F prototype \u002F research \u002F task) with explicit dependencies and an open frontier agents self-assign into. Plans; never executes. Use this skill whenever the user is working with wayfinder, and on phrases like \"chart this effort\", \"break this into tickets\", \"what's the frontier\", \"plan across sessions\", or \"map out the dependencies for this work\".","https:\u002F\u002Fgithub.com\u002FJRichlen\u002Fagent-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fwayfinder\u002Fskills\u002Fwayfinder\u002FSKILL.md",[],[],{"skillNo":66,"name":67,"nameZh":6,"description":68,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":69,"categories":70,"keywords":71,"publishedAt":19},"01M78DCB8APFG6B3G7D34WZ2H9","wave-research","Staged, adversarial research that tries to break a thesis before believing it: waves, checkpoints, falsifiable sub-claims, rival hypotheses, an ACH matrix, source-incentive tags and a verified final thesis. Also carries Workflow templates for a one-pass research-and-verify fan-out. Use to stress-test or vet a contested claim; not for quick lookups.","https:\u002F\u002Fgithub.com\u002FJRichlen\u002Fagent-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fwave-research\u002Fskills\u002Fwave-research\u002FSKILL.md",[],[],{"skillNo":73,"name":74,"nameZh":6,"description":75,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":76,"categories":77,"keywords":78,"publishedAt":19},"01M78DC30012WTD0B2T21ZFVFG","second-opinion","Offer-only validation pipeline for a verdict or recommendation. Use when the user asks to validate, verify, fact-check, or get a second opinion on a claim or recommendation — \"are you sure\", \"double-check this\", \"second opinion\". Batches fact-checking against sources, adds scoped advisor personas for the angles that need judgment, then re-emits the response grouped into verified, flagged, and conflict, with a mandatory validation delta. Requires a subagent-spawning tool; without one it must decline rather than simulate.","https:\u002F\u002Fgithub.com\u002FJRichlen\u002Fagent-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fvoice\u002Fskills\u002Fsecond-opinion\u002FSKILL.md",[],[],{"skillNo":80,"name":81,"nameZh":6,"description":82,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":83,"categories":84,"keywords":85,"publishedAt":19},"01M78DBTTZAK0Z8FZ7RCFX92NM","machine-voice","Compression rules for standalone machine-read artifacts the reader scans — agent traces, tool-call logs, status lines, progress updates, handoff notes, state dumps, config, schemas, reference cards, and structured data another agent parses. Applies three layers: lexical compression, markdown structure as navigation, and emojis as typed status markers. Reach for it when emitting one of those artifacts on its own. Companion to human-voice, which governs prose a human reads and also owns any table or block that serves surrounding explanation. Neither skill applies to code or file contents the user will save, commit messages, creative writing, or turns that are only a clarifying question — those ship unstyled and uncompressed.","https:\u002F\u002Fgithub.com\u002FJRichlen\u002Fagent-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fvoice\u002Fskills\u002Fmachine-voice\u002FSKILL.md",[],[],{"skillNo":87,"name":88,"nameZh":6,"description":89,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":90,"categories":91,"keywords":92,"publishedAt":19},"01M78DBH7MPRD7ZNKVRA6JK76S","human-voice","Style for prose a human reads — answers, recommendations, explanations, research summaries, comparisons. Use on every conversational reply that contains prose for a person to read: triage intent, lead with the verdict, keep it scannable, tag claim confidence. Reach for it whenever answering a technical question, comparing options, or making a recommendation, even when the user says nothing about style. Companion to machine-voice, which instead governs standalone machine-read artifacts — agent traces, tool-call logs, status lines, progress updates, handoff notes, state dumps, config, schemas, reference cards, and structured data another agent parses. Neither skill applies to code or file contents, commit messages, creative writing, or turns that are only a clarifying question — those ship unstyled.","https:\u002F\u002Fgithub.com\u002FJRichlen\u002Fagent-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fvoice\u002Fskills\u002Fhuman-voice\u002FSKILL.md",[],[],{"skillNo":94,"name":95,"nameZh":6,"description":96,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":97,"categories":98,"keywords":99,"publishedAt":19},"01M78DB9587AD3Y46FYVE2V3T4","ai-writing-mistakes","Catch and repair the tells that mark prose as machine-written — the negation-flip (\"it's not X, it's Y\"), rule-of-three padding, hedge stacking, unearned intensifiers, filler verbs like delve and leverage, participle trailers (\"...ensuring seamless integration\"), sycophantic openers, empty closing summaries, and uniform em-dash\u002Fsentence rhythm. Use when drafting or revising any prose a human will read, including documents you were asked to author — READMEs, docs, PR bodies, release notes, emails — and when the user asks whether something \"sounds like AI\" or wants writing tightened. This is a wording pass, not a voice: human-voice still decides layout and confidence tags, machine-voice still owns traces and logs, and neither this skill nor those two touch code, commit messages, or quoted text.","https:\u002F\u002Fgithub.com\u002FJRichlen\u002Fagent-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fvoice\u002Fskills\u002Fai-writing-mistakes\u002FSKILL.md",[],[]]