[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7A4Q1G3B66ASVC5R61C7WMG":3,"skill-same:01M7A4Q1G3B66ASVC5R61C7WMG":69},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":13,"repositoryHtmlUrl":26,"repositoryHomepageUrl":6,"ownerHtmlUrl":27,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":28,"install":29,"resources":42},"01M7A4Q1G3B66ASVC5R61C7WMG","model-routing",null,"Use before the first search or edit of work that needs more than two steps, and whenever bounded work might be handed to another model through `pi-dispatch`. Chooses the owner (main session, `pi-dispatch` worker, or fresh session) and the model slot. Not for single-step requests or questions answerable directly.","jhonDoe15\u002Fskills","jhonDoe15","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F62393042?v=4","01M7A4NRG3NB5KYTSSMMGHHAMC","01M7A4NRGQHAAQA7Q7Q3RRVGAS",0,"MIT","https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fmodel-routing\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: model-routing\ndescription: Use before the first search or edit of work that needs more than two steps, and whenever bounded work might be handed to another model through `pi-dispatch`. Chooses the owner (main session, `pi-dispatch` worker, or fresh session) and the model slot. Not for single-step requests or questions answerable directly.\ndisable-model-invocation: false\n---\n\n# Model Routing\n\nDecide who does each unit of work, and on which model, before doing it. Then\ndispatch with `pi-dispatch` or keep the work here.\n\n## Route first\n\nBefore the first search or edit on work that needs more than two steps, state a\n**route**:\n\n1. Name the intended outcome, done boundary, and overreach boundary.\n2. Separate judgment from mechanical execution.\n3. Assign each unit to the main session, `pi-dispatch`, or a fresh session.\n\nKeeping work in the main session is a route and needs the same decision.\nDefault to delegation when it reduces context load, enables parallel progress,\nor creates a clearer completion boundary. Keep the work here when handoff\noverhead would exceed the bounded work.\n\n## Choose the owner\n\n- **Main session:** work that depends on reasoning already held here. It owns\n  decisions, orchestration, integration, and review.\n- **`pi-dispatch`:** bounded, self-contained work whose handoff the main session\n  will consume.\n- **Fresh session:** a new phase, a clean-slate second opinion, or an unrelated\n  thread. Load `take-it-offline` by its exact canonical name to create the\n  handoff, recommend the slot, then wait for the user to open the session.\n\nEach dispatch owns exactly one bounded task. Its brief supplies the necessary\ninputs, one outcome, allowed scope, constraints, and checkable completion\ncriteria. The task may take several steps; each must advance those criteria.\nMeeting them ends the worker's authority, and adjacent work returns to the\nparent.\n\n## Choose the slot\n\nChoose the family by the shape of the evidence, not by whether the task\ncontains judgment:\n\n- **OpenAI** for closed-world work: the contract, rubric, source set, or\n  expected validation is explicit and coverage can be enumerated. This includes\n  implementation, API and contract research, test planning, static checks, and\n  exact spec comparison.\n- **Anthropic** for open-world work: the worker must infer intent, expose hidden\n  invariants, generate causal hypotheses, reconcile ambiguous evidence, or\n  reason about interactions across state, time, or ownership boundaries.\n- Wide independent items favor OpenAI. A cohesive problem whose parts share\n  implicit behavior favors Anthropic. File count alone selects neither family\n  nor tier.\n\n| Work | Slot |\n|---|---|\n| Deterministic transformation or implementation with exact checks, even across several files | `openai-small` |\n| Contract or API research, test mapping, or rubric-driven spec, standards, and static review | `openai-mid` |\n| Exhaustive closed-world reconciliation where omissions are hard to detect, or proven mid-tier reasoning insufficiency | `openai-top` |\n| Factual summary, eligibility gate, or bounded local semantic check | `anthropic-small` |\n| Unknown-cause debugging, implicit-invariant review, ambiguous intent, or interacting-module judgment | `anthropic-mid` |\n| Unresolved cross-system behavior or adversarial review where a wrong answer is both hard to detect and costly | `anthropic-top` |\n\nStart at `small` and promote only on evidence:\n\n- `small` when a precise brief and deterministic validation can expose a wrong\n  result. Large settled edits may stay small.\n- `mid` when local judgment remains, validation is partly interpretive, or the\n  worker must discover and connect evidence.\n- `top` only when mid-tier reasoning has proved insufficient, multiple\n  unresolved uncertainties interact across boundaries and failure would be\n  silent or expensive, or required completeness is consequential and omissions\n  cannot be checked cheaply. A sensitive domain alone does not require `top`;\n  sensitivity plus unresolved judgment does.\n\nCalibration: a multi-file refactor with an exact analyzer and tests is\n`openai-small`; a pinned SDK contract audit is `openai-mid`; a stale async-state\nrace is `anthropic-mid`; an unresolved authorization design spanning systems is\n`anthropic-top`.\n\nPromote a wide enumeration only when omissions cannot be checked cheaply.\nEscalate after failure only when the failure reveals missing judgment or\ninsufficient reasoning; fix the brief or tooling at the same tier for execution\nfailures.\n\n## Dispatch\n\nRun `pi-dispatch --list` for the current slots, models, and effort levels; the\ncommand is their only authority. Dispatch from the working directory the worker\nshould use:\n\n```bash\npi-dispatch --list\npi-dispatch \u003Cslot> \u003Cslug> \"\u003Cbounded brief>\"\npi-dispatch \u003Cslot> \u003Cslug> --context \u003Chandoff> \"\u003Cangle-specific brief>\"\npi-dispatch --recursive \u003Cslot> \u003Cslug> [--context \u003Chandoff>] \"\u003Cbounded parent brief>\"\n```\n\nThe command attaches the family and tier prompting profiles itself, from\n`~\u002F.cursor\u002Frules\u002Freferences\u002Fpi-dispatch-prompting.md` unless\n`PI_DISPATCH_PROMPTING_GUIDE` names another file. When the command exits with\n`Prompting guide must be a readable, nonempty file`, set\n`PI_DISPATCH_PROMPTING_GUIDE` to the absolute path of\n[`references\u002Fpi-dispatch-prompting.md`](references\u002Fpi-dispatch-prompting.md)\nin this skill and rerun. Keep only task-specific judgment in the brief; do not\ncopy or weaken those profiles. It\nalso attaches the optional `--context` handoff without copying it into the\ncommand, and prints only the output handoff path. Output metadata records the\ncontext path and hash when supplied.\n\nWhen several workers share context, load `take-it-offline` by its exact\ncanonical name once to create the handoff, and reuse its path with a small brief\nfor each angle. Without shared context, pass the\ncomplete brief directly or through stdin.\n\nEvery ordinary dispatch is a depth-one **bounded worker**. Its authority is the\nbrief; shared context supplies facts rather than scope. Completion ends the\nrun.\n\nAfter a dispatch returns, read its handoff, inspect any diff, and verify the\ncritical path before accepting the work.\n\n## Opt into recursion\n\nOnly the main session may pass `--recursive`; the command rejects it from a\ndispatched agent. The resulting bounded parent may route settled child units\nthrough `pi-dispatch`:\n\n- It keeps judgment that depends on its current reasoning.\n- It creates one shared handoff for common child context and varies each\n  child's brief by angle.\n- Each child gets one outcome, a checkable stop condition, and exclusive\n  ownership of any files it writes.\n- It consumes and verifies child handoffs before consolidating.\n- It carries child results and handoff paths upward and keeps their working\n  context at the child level.\n\nOpted-in roots are limited to one child level and three children. Child\nhandoffs nest under the parent's dispatch directory and are listed in the\nparent handoff metadata. The main session receives the root handoff and follows\nchild paths only when verification requires it.\n\n## Re-route and report\n\nRe-route only at a boundary: exploration finished, a coherent patch landed, or\nvalidation changed the diagnosis. Report each dispatched unit with its slot and\nrecord any misroute.\n\n## Failure behavior\n\n- `pi-dispatch` is not found (`command -v pi-dispatch` fails): route every unit\n  to the main session or a fresh session, and report that dispatch was\n  unavailable.\n- A slot's model is unavailable before inference: record an availability\n  failure and reroute at the equivalent tier by task shape. Do not treat it as\n  a quality result.\n- A dispatch exits nonzero or returns no handoff: report the exit status and\n  stderr, and do not treat the unit as done.\n- `take-it-offline` is unavailable when a shared handoff or fresh session is\n  needed: stop that branch with\n  `Missing internal dependency \"take-it-offline\"`. Do not substitute a\n  handoff written from local guidance.\n\n## Completion\n\nComplete when:\n\n- a route was stated before the first search or edit, with every unit assigned\n  an owner;\n- every dispatched unit has a slot chosen by evidence shape and tier, and a\n  brief with one outcome, allowed scope, and checkable completion criteria;\n- every returned handoff was read, and its diff and critical path were\n  verified before acceptance;\n- each re-route happened at a boundary and each misroute was recorded.\n","# Model Routing\n\nDecide who does each unit of work, and on which model, before doing it. Then\ndispatch with `pi-dispatch` or keep the work here.\n\n## Route first\n\nBefore the first search or edit on work that needs more than two steps, state a\n**route**:\n\n1. Name the intended outcome, done boundary, and overreach boundary.\n2. Separate judgment from mechanical execution.\n3. Assign each unit to the main session, `pi-dispatch`, or a fresh session.\n\nKeeping work in the main session is a route and needs the same decision.\nDefault to delegation when it reduces context load, enables parallel progress,\nor creates a clearer completion boundary. Keep the work here when handoff\noverhead would exceed the bounded work.\n\n## Choose the owner\n\n- **Main session:** work that depends on reasoning already held here. It owns\n  decisions, orchestration, integration, and review.\n- **`pi-dispatch`:** bounded, self-contained work whose handoff the main session\n  will consume.\n- **Fresh session:** a new phase, a clean-slate second opinion, or an unrelated\n  thread. Load `take-it-offline` by its exact canonical name to create the\n  handoff, recommend the slot, then wait for the user to open the session.\n\nEach dispatch owns exactly one bounded task. Its brief supplies the necessary\ninputs, one outcome, allowed scope, constraints, and checkable completion\ncriteria. The task may take several steps; each must advance those criteria.\nMeeting them ends the worker's authority, and adjacent work returns to the\nparent.\n\n## Choose the slot\n\nChoose the family by the shape of the evidence, not by whether the task\ncontains judgment:\n\n- **OpenAI** for closed-world work: the contract, rubric, source set, or\n  expected validation is explicit and coverage can be enumerated. This includes\n  implementation, API and contract research, test planning, static checks, and\n  exact spec comparison.\n- **Anthropic** for open-world work: the worker must infer intent, expose hidden\n  invariants, generate causal hypotheses, reconcile ambiguous evidence, or\n  reason about interactions across state, time, or ownership boundaries.\n- Wide independent items favor OpenAI. A cohesive problem whose parts share\n  implicit behavior favors Anthropic. File count alone selects neither family\n  nor tier.\n\n| Work | Slot |\n|---|---|\n| Deterministic transformation or implementation with exact checks, even across several files | `openai-small` |\n| Contract or API research, test mapping, or rubric-driven spec, standards, and static review | `openai-mid` |\n| Exhaustive closed-world reconciliation where omissions are hard to detect, or proven mid-tier reasoning insufficiency | `openai-top` |\n| Factual summary, eligibility gate, or bounded local semantic check | `anthropic-small` |\n| Unknown-cause debugging, implicit-invariant review, ambiguous intent, or interacting-module judgment | `anthropic-mid` |\n| Unresolved cross-system behavior or adversarial review where a wrong answer is both hard to detect and costly | `anthropic-top` |\n\nStart at `small` and promote only on evidence:\n\n- `small` when a precise brief and deterministic validation can expose a wrong\n  result. Large settled edits may stay small.\n- `mid` when local judgment remains, validation is partly interpretive, or the\n  worker must discover and connect evidence.\n- `top` only when mid-tier reasoning has proved insufficient, multiple\n  unresolved uncertainties interact across boundaries and failure would be\n  silent or expensive, or required completeness is consequential and omissions\n  cannot be checked cheaply. A sensitive domain alone does not require `top`;\n  sensitivity plus unresolved judgment does.\n\nCalibration: a multi-file refactor with an exact analyzer and tests is\n`openai-small`; a pinned SDK contract audit is `openai-mid`; a stale async-state\nrace is `anthropic-mid`; an unresolved authorization design spanning systems is\n`anthropic-top`.\n\nPromote a wide enumeration only when omissions cannot be checked cheaply.\nEscalate after failure only when the failure reveals missing judgment or\ninsufficient reasoning; fix the brief or tooling at the same tier for execution\nfailures.\n\n## Dispatch\n\nRun `pi-dispatch --list` for the current slots, models, and effort levels; the\ncommand is their only authority. Dispatch from the working directory the worker\nshould use:\n\n```bash\npi-dispatch --list\npi-dispatch \u003Cslot> \u003Cslug> \"\u003Cbounded brief>\"\npi-dispatch \u003Cslot> \u003Cslug> --context \u003Chandoff> \"\u003Cangle-specific brief>\"\npi-dispatch --recursive \u003Cslot> \u003Cslug> [--context \u003Chandoff>] \"\u003Cbounded parent brief>\"\n```\n\nThe command attaches the family and tier prompting profiles itself, from\n`~\u002F.cursor\u002Frules\u002Freferences\u002Fpi-dispatch-prompting.md` unless\n`PI_DISPATCH_PROMPTING_GUIDE` names another file. When the command exits with\n`Prompting guide must be a readable, nonempty file`, set\n`PI_DISPATCH_PROMPTING_GUIDE` to the absolute path of\n[`references\u002Fpi-dispatch-prompting.md`](references\u002Fpi-dispatch-prompting.md)\nin this skill and rerun. Keep only task-specific judgment in the brief; do not\ncopy or weaken those profiles. It\nalso attaches the optional `--context` handoff without copying it into the\ncommand, and prints only the output handoff path. Output metadata records the\ncontext path and hash when supplied.\n\nWhen several workers share context, load `take-it-offline` by its exact\ncanonical name once to create the handoff, and reuse its path with a small brief\nfor each angle. Without shared context, pass the\ncomplete brief directly or through stdin.\n\nEvery ordinary dispatch is a depth-one **bounded worker**. Its authority is the\nbrief; shared context supplies facts rather than scope. Completion ends the\nrun.\n\nAfter a dispatch returns, read its handoff, inspect any diff, and verify the\ncritical path before accepting the work.\n\n## Opt into recursion\n\nOnly the main session may pass `--recursive`; the command rejects it from a\ndispatched agent. The resulting bounded parent may route settled child units\nthrough `pi-dispatch`:\n\n- It keeps judgment that depends on its current reasoning.\n- It creates one shared handoff for common child context and varies each\n  child's brief by angle.\n- Each child gets one outcome, a checkable stop condition, and exclusive\n  ownership of any files it writes.\n- It consumes and verifies child handoffs before consolidating.\n- It carries child results and handoff paths upward and keeps their working\n  context at the child level.\n\nOpted-in roots are limited to one child level and three children. Child\nhandoffs nest under the parent's dispatch directory and are listed in the\nparent handoff metadata. The main session receives the root handoff and follows\nchild paths only when verification requires it.\n\n## Re-route and report\n\nRe-route only at a boundary: exploration finished, a coherent patch landed, or\nvalidation changed the diagnosis. Report each dispatched unit with its slot and\nrecord any misroute.\n\n## Failure behavior\n\n- `pi-dispatch` is not found (`command -v pi-dispatch` fails): route every unit\n  to the main session or a fresh session, and report that dispatch was\n  unavailable.\n- A slot's model is unavailable before inference: record an availability\n  failure and reroute at the equivalent tier by task shape. Do not treat it as\n  a quality result.\n- A dispatch exits nonzero or returns no handoff: report the exit status and\n  stderr, and do not treat the unit as done.\n- `take-it-offline` is unavailable when a shared handoff or fresh session is\n  needed: stop that branch with\n  `Missing internal dependency \"take-it-offline\"`. Do not substitute a\n  handoff written from local guidance.\n\n## Completion\n\nComplete when:\n\n- a route was stated before the first search or edit, with every unit assigned\n  an owner;\n- every dispatched unit has a slot chosen by evidence shape and tier, and a\n  brief with one outcome, allowed scope, and checkable completion criteria;\n- every returned handoff was read, and its diff and critical path were\n  verified before acceptance;\n- each re-route happened at a boundary and each misroute was recorded.\n",{"name":5,"description":7,"disable-model-invocation":19},"main","00b68ded1da612384ffdf1f27cf8228b9c2d4428","skills\u002Fmodel-routing","https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills","https:\u002F\u002Fgithub.com\u002FjhonDoe15","2026-10-16 01:12:43",{"source":8,"skillName":5,"commands":30,"prompt":40,"zipAvailable":41},[31,34,37],{"tool":32,"command":33},"npx","npx skills add jhonDoe15\u002Fskills --skill model-routing",{"tool":35,"command":36},"bunx","bunx skills add jhonDoe15\u002Fskills --skill model-routing",{"tool":38,"command":39},"pnpm","pnpm dlx skills add jhonDoe15\u002Fskills --skill model-routing","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fmodel-routing\u002FSKILL.md",true,[43,49,53,57,61,65],{"path":44,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":47,"sourceHtmlUrl":48},"evals\u002Fcomponent-take-it-offline.json","REFERENCE","参考文档",2812,"https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fmodel-routing\u002Fevals\u002Fcomponent-take-it-offline.json",{"path":50,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":51,"sourceHtmlUrl":52},"evals\u002Foutcome.json",3200,"https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fmodel-routing\u002Fevals\u002Foutcome.json",{"path":54,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":55,"sourceHtmlUrl":56},"evals\u002Fpackage-closure.json",552,"https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fmodel-routing\u002Fevals\u002Fpackage-closure.json",{"path":58,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":59,"sourceHtmlUrl":60},"evals\u002Frole.json",7516,"https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fmodel-routing\u002Fevals\u002Frole.json",{"path":62,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":63,"sourceHtmlUrl":64},"evals\u002Ftrigger.json",3814,"https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fmodel-routing\u002Fevals\u002Ftrigger.json",{"path":66,"resourceType":45,"resourceTypeDesc":46,"sizeBytes":67,"sourceHtmlUrl":68},"references\u002Fpi-dispatch-prompting.md",3804,"https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fmodel-routing\u002Freferences\u002Fpi-dispatch-prompting.md",[70,77,84,91,98,105],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":18},"01M7A4R91YWB9YCB90RYMEKP5Y","writing-foundation","Shared writing behavior consumed by to-humans and agent-writing; not a user-facing outcome.","https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fwriting-foundation\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":18},"01M7A4R5WK4J9HGENYK6JZN124","to-humans","Audience outcome for ordinary replies, progress updates, and requested prose whose primary reader is a person. Select it independently for human-facing expression, including alongside one substantive Primary outcome. Excludes agent-facing artifacts, fresh-context handoffs, Agent Skill packages, and non-prose work products.","https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fto-humans\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":18},"01M7A4R3DVDG3FN21JN4VAXXRF","ticket-scope","Private per-candidate scope judgment consumed by Slice Plan and PR Carver; not a user-facing planning outcome.","https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fticket-scope\u002FSKILL.md",[],[],{"skillNo":92,"name":93,"nameZh":6,"description":94,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":95,"categories":96,"keywords":97,"publishedAt":18},"01M7A4QZNMGMB448HXQQ052EYH","take-ticket","Use when a settled ticket must reach a reviewed-ticket result. Used directly or by Dispatch Work. Excludes implementation-only work, review-only work, ticket topology, publication, release, and multi-ticket dispatch.","https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Ftake-ticket\u002FSKILL.md",[],[],{"skillNo":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":102,"categories":103,"keywords":104,"publishedAt":18},"01M7A4QX88C6QAFMWT7VWTYJXS","take-it-offline","Use when carrying current work into a fresh agent context through a continuation handoff. Excludes human summaries, archival specifications, general agent-facing artifacts, and Agent Skill authoring.","https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Ftake-it-offline\u002FSKILL.md",[],[],{"skillNo":106,"name":107,"nameZh":6,"description":108,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":109,"categories":110,"keywords":111,"publishedAt":18},"01M7A4QVFT7BVQW0KA651KK6BF","slice-plan","Private set-level decomposition consumed by Carve; not a user-facing planning outcome.","https:\u002F\u002Fgithub.com\u002FjhonDoe15\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fslice-plan\u002FSKILL.md",[],[]]