[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M79DSM2B1RGQJMX0SPKPPD5W":3,"skill-same:01M79DSM2B1RGQJMX0SPKPPD5W":144},{"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":30,"updatedAt":31,"install":32,"resources":45},"01M79DSM2B1RGQJMX0SPKPPD5W","working-backwards",null,"Define a software feature through PRFAQ and staged documents when the user requests working backwards before coding.","AO-HyS\u002Fdevelopment-system","AO-HyS","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F286554351?v=4","01M79DCXZGC6076EH5QATNXCY2","AO HyS","01M79DCY06TSBCJ946VP0BT0RG",0,"MIT","https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: working-backwards\ndescription: Define a software feature through PRFAQ and staged documents when the user requests working backwards before coding.\n---\n\nFor a standalone engineering report rather than a workflow artifact, read\n`report-reference.md` and use the shared report helper. Continue below for a\nWorking Backwards initiative.\n\n# Working Backwards in T3 Code\n\nStart with a non-technical Product Grill With Docs, then work backwards from the finished customer experience. Accept an ordinary sentence; never require a structured mega-prompt. T3 Code is the conversational and execution surface. HumanLayer is not required.\n\n## Create the workspace\n\nResolve the Git repository root, normalized remote, and exact `git rev-parse HEAD`. Derive short path-safe repository and feature slugs. Use the repository's configured private planning root when one exists; otherwise create outside Git under the operator HOME:\n\n```text\n~\u002F.development-system\u002Fprivate\u002Fworking-backwards\u002F\u003Crepository-slug>-\u003Cfeature-slug>\u002F\n```\n\nUse that exact feature-specific directory as `workspaceDir`. Never place this workspace in a synchronized, published, or Git-tracked path. Markdown is the canonical agent-readable source. The helper derives a plain Reader model from Markdown plus workflow-state JSON, regenerates one current `\u003Cinitiative-slug>.html`, and renders every available artifact under `reader-history\u002F` with the same visual system. Do not author feature-specific HTML, Canvas, Agent Native, or Integrar artifacts.\n\nRun the installed helper before acting:\n\n```text\nnode \u003Cskill-dir>\u002Fscripts\u002Ft3-workflow.mjs status \\\n  --workspace-dir \u003Cabsolute-workspace-dir> \\\n  --repository-path \u003Cabsolute-repository-root> \\\n  --repository-identity \u003Cremote> \\\n  --repository-revision \u003Crevision>\n```\n\nIts `currentPhase` and `action` are authoritative. After creating or revising an artifact, regenerate the Reader:\n\n```text\nnode \u003Cskill-dir>\u002Fscripts\u002Ft3-workflow.mjs render \\\n  --workspace-dir \u003Cabsolute-workspace-dir> \\\n  --repository-path \u003Cabsolute-repository-root> \\\n  --repository-identity \u003Cremote> \\\n  --repository-revision \u003Crevision>\n```\n\nAfter each substantive render, start or refresh the bounded local review surface:\n\n```text\nnode \u003Cskill-dir>\u002Fscripts\u002Freader-live.mjs \\\n  --workspace \u003Cabsolute-workspace-dir> \\\n  --reader \u003Cinitiative-slug>.html \\\n  --ttl-minutes 120\n```\n\nReturn its localhost URL, the Reader file link, and the active Markdown file. Keep the yielded process\u002Fsession so it can be stopped when the user finishes reviewing. If the user says they are away, on another computer, or otherwise cannot reach localhost, add `--tunnel`. Return the tokenized temporary URL and its expiry. This is explicit remote-review authorization for that one private workspace only; it is not authorization to publish, index, sync, or retain planning content. Stop the process when the user says they reviewed it, when the workflow is abandoned, or at TTL expiry.\n\n## Run one document at a time\n\nUse exactly this order for the Standard profile:\n\n1. `product-grill`: Product Grill With Docs by Topic. Settle actor, problem, desired outcome, future experience, boundaries, and expectations. Do not design entities, interfaces, storage, or architecture here.\n2. `customer-story`: a compact, non-technical future customer narrative derived from the approved Product Grill.\n3. `technical-grill`: Technical Grill With Docs driven by the approved story, selected profile, repository evidence, and risk triggers. Settle behavior, entities, states, interfaces, data, security, testing, rollout, and technical trade-offs without repeating product Topics.\n4. `research-questions`: only questions about current behavior, code, users, or external facts still unresolved after the Technical Grill.\n5. `research-report`: live code\u002Fruntime and primary-source answers, with facts, inferences, and unknowns separated.\n6. `product-contract`: observable behavior, scope, success, states, permissions, errors, recovery, compatibility, and rejected product options.\n7. `technical-contract`: entities, invariants, interfaces, reads, writes, events, migrations, security, rollback, test seams, and rejected technical options.\n8. `implementation-map`: narrow vertical tickets with outcome, acceptance, checks, dependencies, and one truthful executable frontier.\n9. `t3-handoff`: compact private handoff bound to the approved artifacts and exact first slice; `implementationAuthorized: false`.\n\nCreate numbered Markdown files with these stems:\n\n```text\n01-product-grill.md\n02-customer-story.md\n03-technical-grill.md\n04-research-questions.md\n05-research-report.md\n06-product-contract.md\n07-technical-contract.md\n08-implementation-map.md\n```\n\nEvery canonical artifact starts with:\n\n```yaml\n---\nworking_backwards_role: \u003Crole>\nworking_backwards_status: draft\ntitle: \u003Cshort document title>\nsummary: \u003Cone or two sentence settled summary>\n---\n```\n\nThe implementation map also declares `working_backwards_first_slice: \u003Cticket-id>` and contains that exact ticket with satisfied dependencies and an executable acceptance path.\n\nThe helper's first-slice field identifies the first executable frontier. It\ndoes not reduce a later instruction to implement the spec to that one ticket.\nThe spec links every ticket, and governed implementation retrieves and tracks\nall authorized tickets through their required evidence. Current and retained\nuser authorization determines the endpoint across threads.\n\nCreate or revise exactly one artifact at a time. Begin from the user's ordinary sentence and draft the first useful version yourself. Ask only when an unresolved decision can change the active document; inspect repository facts and use settled conversation context instead of asking the user to repeat discoverable information. Write every settled answer into that Markdown, keep the document concise, and regenerate `\u003Cinitiative-slug>.html`. Remain on the active document until the user clearly approves it.\n\n### Ask by Topic\n\nKeep one Topic per turn. Within that Topic, batch `1–3` mutually related decisions when answering them together reduces back-and-forth. Ask only one when only one meaningful decision remains. Never combine different Topics merely to fill the tool.\n\nCompose the round through the installed helper:\n\n```text\nnode \u003Cskill-dir>\u002Fscripts\u002Ftopic-questions.mjs \\\n  --input \u003Cabsolute-topic-json> \\\n  --format both\n```\n\nThe input contains one `topic`, optional `settledContext`, and `1–3` `decisions`. Each decision has a stable snake-case `id`, a short header, the question, and `2–3` mutually exclusive options. Each option supplies a short label, its impact or trade-off, a concrete example, and exactly one option is marked `recommended: true`. The helper orders the recommendation first, labels it `(Recommended)`, and emits both interfaces from the same source.\n\n- When `request_user_input` is available in the active collaboration mode, call it with the helper's exact `native` payload.\n- When it is unavailable, send the helper's exact `chat` fallback. Do not improvise a different set of options.\n- The client supplies a free-form Other choice for the native interface; the chat fallback explicitly accepts a different option or added nuance.\n- After the reply, persist the selected decisions, rationale, changed assumptions, and remaining unknowns in the active Markdown before moving to the next Topic.\n- If the answer changes the Topic map, add, remove, or reorder later Topics openly. Do not silently preserve stale questions.\n\nThe helper is a deterministic composer, not a source of product judgment. The agent still chooses the relevant Topic, recommendation, trade-offs, and examples from current evidence.\n\nWrite settled plans, not transcripts. Prefer a short summary followed by `3–8` descriptive sections. Keep the evidence needed to understand or approve the decision; move raw logs and exhaustive source notes to referenced evidence files. Never compress away scope, acceptance criteria, risks, decisions, dependencies, or unknowns.\n\nFor a broad initiative with named workstreams, preserve those workstream names as a coverage ledger in every later artifact. Each workstream must be marked `covered`, `unchanged`, `deferred with reason`, or `blocked with evidence`. A Research Report answers residual questions; it never silently replaces the broader Product Grill, story, or convergence scope. The Reader's current page must therefore remain understandable as one phase of the whole initiative.\n\nOn every user reply while `action: review-artifact`, pass the entire message to:\n\n```text\nnode \u003Cskill-dir>\u002Fscripts\u002Ft3-workflow.mjs turn \\\n  --workspace-dir \u003Cabsolute-workspace-dir> \\\n  --repository-path \u003Cabsolute-repository-root> \\\n  --message \u003Centire-user-message> \\\n  --repository-identity \u003Cremote> \\\n  --repository-revision \u003Crevision>\n```\n\nOnly `approval.accepted: true` advances. Ordinary natural-language approval is sufficient for the one active artifact; the user never has to memorize a phrase. A direct affirmation such as `está muy bien`, `me gusta`, `todo correcto`, `perfecto todo aquí`, or `sí` is sufficient. So is a clear no-change statement paired with normal movement language, for example `no tengo cambios, sigue`, `por mí está bien, vamos a la siguiente fase`, `ya quedó, dale`, or `continúa con el siguiente documento`. Feedback revises the same file. Questions, negation, ambiguity, reported approval, combined gates, or a reply that also asks for changes never advance. When affirmation and requested changes coexist, feedback wins: revise the artifact and present its new bytes for approval again.\n\nProduct-convergence prompts may explicitly opt into a reference pack and ask the Technical Grill to infer architecture choices that repository evidence settles. That is prompt-local behavior, not the default for ordinary or complex Working Backwards work.\n\n### Architecture-convergence program\n\nWhen an opted-in product-convergence prompt requests a repository-wide architecture program, add `working_backwards_program: architecture-convergence` to the Technical Grill frontmatter and load the canonical architecture reference pack. The Grill must include the reference pack's complete `Dimension | Current evidence | Decision | Enforcement` matrix for all of these identifiers:\n\n`repository-map`, `module-boundaries`, `dependency-direction`, `file-placement`, `frontend-composition`, `component-design`, `backend-contracts`, `type-contracts`, `testing-strategy`, `documentation`, `performance-security`, `observability`, and `migration-sequencing`.\n\nDo not infer that a repository is understandable merely because its top-level apps have names. Inspect representative paths, imports, public Interfaces, runtime boundaries, tests, docs, generated code, compatibility surfaces, provider adapters, and automation. For components and modules, judge cohesion, responsibility, state ownership, dependency direction, Interface depth, and reuse by proven invariant; never invent a universal line-count limit. For Convex, cover queries, mutations, actions, validators, authorization, indexes, pagination, bounded reads, subscriptions, contention, scheduling, storage, maintained components, and provider adapters.\n\nAgent guardrails, global anti-slop policy, and Release Train design belong to the Development System. Product convergence may verify the installed repository adapter and the product commands it declares, but it must not create product migration workstreams or tickets to redesign those capabilities. Product-specific documentation still explains domain ownership and file\u002Ftest locality; release execution evidence may appear in the final delivery report without making Release Train part of the architecture migration.\n\nAn incomplete matrix fails closed at Technical Grill approval. A row marked `unproven` creates a focused research obligation. Every `change`, `remove`, and unresolved `unproven` row must remain traceable through Product Contract, Technical Contract, and Implementation Map to an executable ticket, an evidence-backed no-change conclusion, or a blocking human gate. Research about one provider or hotspot never replaces the repository-wide architecture program.\n\nAfter an approval, say plainly:\n\n> Terminamos **\u003Cdocumento>**. Siguiente: **\u003Cdocumento siguiente>**.\n\nThen create only the next artifact. Draft it from the approved evidence already available and ask its first high-leverage question only if one remains. Do not dump the remaining process into chat.\n\n## Use the shared technical Reader\n\nThe generated `\u003Cinitiative-slug>.html` is the current-phase entrypoint. Every available prior artifact also has a full Reader page under `reader-history\u002F`; the artifact rail links to those rendered pages, not raw Markdown. Every page states `Fase X de 9` and explains that it is one part of the wider initiative. The format is static, escaped, responsive, offline-capable, and `noindex`. It shows:\n\n- a compact artifact\u002Fphase rail and source links;\n- one continuous `760–900px` technical document;\n- an active `On this page` outline from descriptive headings;\n- restrained title, summary, status, priority, profile, reading time, dates, and repository metadata;\n- the next action and exact human gate in plain language;\n- first-class diagrams, code, tables, decisions, risks, testing, and rollout evidence;\n- `implementationAuthorized: false` until a separate Implement Preview.\n\nOn narrow screens, the document comes first and artifact\u002Foutline navigation becomes secondary controls. The Reader model is plain JSON derived from canonical Markdown and workflow state. Other local development-system surfaces may reuse the same renderer by supplying that model; Working Backwards does not create another presentation system.\n\nUse the Markdown itself to request richer review blocks. They remain canonical\nsource and the shared renderer turns them into safe local visuals:\n\n````markdown\n```mermaid\nflowchart LR\n  A[Idea] --> B{Contrato claro}\n  B -->|Sí| C[Implementación]\n```\n\n```chart\n{\"type\":\"bar\",\"title\":\"Confianza por fase\",\"labels\":[\"Historia\",\"Contrato\",\"Tickets\"],\"values\":[35,72,94]}\n```\n\n```typescript\nexport const outcome = \"visible\";\n```\n````\n\nThe renderer embeds the pinned official Mermaid runtime and supports its diagram families, including flowchart, Gantt\u002Fprogress, sequence, timeline, and architecture, without a home-grown parser. Diagrams provide inline, expanded, and fullscreen reading states; drag\u002Fpan, wheel or trackpad movement, pinch, zoom, fit, reset, copy, and source inspection. Fit preserves a readable Mermaid scale; wide diagrams use the available viewport and pan or scroll horizontally instead of shrinking labels into an unreadable overview. Local bar or line charts use explicit JSON and retain an accessible data table. Semantic Markdown tables, sparse callouts, and fenced code\u002Fdiff blocks preserve filename, language, copy, optional line numbers, highlights, wrapping, and clean horizontal scrolling. Prefer the diagram family that matches the relationship and use a chart only for real comparative or sequential data. Never fabricate metrics.\n\nWhen a final implementation report is requested, make it review-complete rather than a link index. Render the full text or clearly separated Reader sections for every product-convergence prompt or handoff the user explicitly asked to review. Include a **Known issues** table with one row per issue, a disposition of `fixed`, `remaining`, or `authorization-blocked`, and concrete evidence. Links may support those sections but never replace their reviewable content.\n\nKeep the Reader self-contained. Do not add CDN, remote fonts, analytics, hosted rendering, or runtime network dependencies. Never send private planning content outside the operator machine except through the bounded, tokenized, expiring tunnel above when the user has indicated remote review. Do not polish per-feature HTML: improve the shared renderer when the default format needs to change.\n\n## Preserve the three gates\n\nThe first three documents use document approval. Hold exactly three formal definition gates:\n\n- `product-contract` -> `approve-product-contract`\n- `technical-contract` -> `approve-technical-contract`\n- `implementation-map` -> `approve-implementation-map`\n\nThe definition helper persists canonical, hash-bound receipts privately under\n`~\u002F.development-system\u002Fprivate\u002Fworking-backwards\u002F\u003Cworkflow-id>\u002F`, binding the\nabsolute root, Git revision and first executable slice. Its legacy Implement\nPreview operation still accepts only that exact first slice and rejects stale\nor concurrent state. That legacy operation is not the execution endpoint for a\nwhole-spec request. A later authorized whole-feature implementation creates a\ngovernance run containing every linked ticket, retains the definition receipts\nas sources and revalidates current repository facts before planning. Drift does\nnot silently discard tickets or replace the user's retained authorization.\n\nAt `create-private-handoff`, write the final handoff only to the helper's `privateHandoffPath`. Populate its required binding frontmatter from `workflowId`, `gateReceiptPath`, repository evidence, all three receipt hashes, the implementation-map hash, and the exact first slice. Run `render` again. Only `action: handoff-ready` proves the handoff matches the approved planning state.\n\n## Stop before implementation\n\nDefault to Standard. Use Quick only for settled, narrow, reversible behavior on one surface. Use Complex for invariants, authorization, sensitive data, destructive behavior, migration\u002Fbackfill, paid activation, uncertain providers, multiple repositories, or difficult rollback.\n\nThis workflow defines work; it does not execute it. Do not implement, publish tracker tickets, commit, push, open a PR, merge, release, deploy, spend money, or perform destructive operations. When `handoff-ready`, tell the user:\n\n> Working Backwards terminó. El spec enlaza todos los tickets y conserva los criterios de aceptación. En una sesión nueva puedes pedir implementar ese spec completo hasta su endpoint autorizado.\n\n## Governed spec and ticket handoff\n\nUse [the shared definition contract](..\u002Fcoding-orchestration\u002Freferences\u002Fspec-ticket-contract.md)\nfor stable criteria, linked tickets, dependencies, rules and required evidence.\nAstra XHigh authors definition and the implementation plan; a different fresh\nAstra XHigh reviews that plan. Fast researchers supply bounded source facts.\nWhen implementation is authorized, use coding-orchestration and its governed\nJev recipe across the complete feature, including nonvisual behavior. Retrieve\nall tickets linked by the spec; preserve settled decisions across threads.\n","For a standalone engineering report rather than a workflow artifact, read\n`report-reference.md` and use the shared report helper. Continue below for a\nWorking Backwards initiative.\n\n# Working Backwards in T3 Code\n\nStart with a non-technical Product Grill With Docs, then work backwards from the finished customer experience. Accept an ordinary sentence; never require a structured mega-prompt. T3 Code is the conversational and execution surface. HumanLayer is not required.\n\n## Create the workspace\n\nResolve the Git repository root, normalized remote, and exact `git rev-parse HEAD`. Derive short path-safe repository and feature slugs. Use the repository's configured private planning root when one exists; otherwise create outside Git under the operator HOME:\n\n```text\n~\u002F.development-system\u002Fprivate\u002Fworking-backwards\u002F\u003Crepository-slug>-\u003Cfeature-slug>\u002F\n```\n\nUse that exact feature-specific directory as `workspaceDir`. Never place this workspace in a synchronized, published, or Git-tracked path. Markdown is the canonical agent-readable source. The helper derives a plain Reader model from Markdown plus workflow-state JSON, regenerates one current `\u003Cinitiative-slug>.html`, and renders every available artifact under `reader-history\u002F` with the same visual system. Do not author feature-specific HTML, Canvas, Agent Native, or Integrar artifacts.\n\nRun the installed helper before acting:\n\n```text\nnode \u003Cskill-dir>\u002Fscripts\u002Ft3-workflow.mjs status \\\n  --workspace-dir \u003Cabsolute-workspace-dir> \\\n  --repository-path \u003Cabsolute-repository-root> \\\n  --repository-identity \u003Cremote> \\\n  --repository-revision \u003Crevision>\n```\n\nIts `currentPhase` and `action` are authoritative. After creating or revising an artifact, regenerate the Reader:\n\n```text\nnode \u003Cskill-dir>\u002Fscripts\u002Ft3-workflow.mjs render \\\n  --workspace-dir \u003Cabsolute-workspace-dir> \\\n  --repository-path \u003Cabsolute-repository-root> \\\n  --repository-identity \u003Cremote> \\\n  --repository-revision \u003Crevision>\n```\n\nAfter each substantive render, start or refresh the bounded local review surface:\n\n```text\nnode \u003Cskill-dir>\u002Fscripts\u002Freader-live.mjs \\\n  --workspace \u003Cabsolute-workspace-dir> \\\n  --reader \u003Cinitiative-slug>.html \\\n  --ttl-minutes 120\n```\n\nReturn its localhost URL, the Reader file link, and the active Markdown file. Keep the yielded process\u002Fsession so it can be stopped when the user finishes reviewing. If the user says they are away, on another computer, or otherwise cannot reach localhost, add `--tunnel`. Return the tokenized temporary URL and its expiry. This is explicit remote-review authorization for that one private workspace only; it is not authorization to publish, index, sync, or retain planning content. Stop the process when the user says they reviewed it, when the workflow is abandoned, or at TTL expiry.\n\n## Run one document at a time\n\nUse exactly this order for the Standard profile:\n\n1. `product-grill`: Product Grill With Docs by Topic. Settle actor, problem, desired outcome, future experience, boundaries, and expectations. Do not design entities, interfaces, storage, or architecture here.\n2. `customer-story`: a compact, non-technical future customer narrative derived from the approved Product Grill.\n3. `technical-grill`: Technical Grill With Docs driven by the approved story, selected profile, repository evidence, and risk triggers. Settle behavior, entities, states, interfaces, data, security, testing, rollout, and technical trade-offs without repeating product Topics.\n4. `research-questions`: only questions about current behavior, code, users, or external facts still unresolved after the Technical Grill.\n5. `research-report`: live code\u002Fruntime and primary-source answers, with facts, inferences, and unknowns separated.\n6. `product-contract`: observable behavior, scope, success, states, permissions, errors, recovery, compatibility, and rejected product options.\n7. `technical-contract`: entities, invariants, interfaces, reads, writes, events, migrations, security, rollback, test seams, and rejected technical options.\n8. `implementation-map`: narrow vertical tickets with outcome, acceptance, checks, dependencies, and one truthful executable frontier.\n9. `t3-handoff`: compact private handoff bound to the approved artifacts and exact first slice; `implementationAuthorized: false`.\n\nCreate numbered Markdown files with these stems:\n\n```text\n01-product-grill.md\n02-customer-story.md\n03-technical-grill.md\n04-research-questions.md\n05-research-report.md\n06-product-contract.md\n07-technical-contract.md\n08-implementation-map.md\n```\n\nEvery canonical artifact starts with:\n\n```yaml\n---\nworking_backwards_role: \u003Crole>\nworking_backwards_status: draft\ntitle: \u003Cshort document title>\nsummary: \u003Cone or two sentence settled summary>\n---\n```\n\nThe implementation map also declares `working_backwards_first_slice: \u003Cticket-id>` and contains that exact ticket with satisfied dependencies and an executable acceptance path.\n\nThe helper's first-slice field identifies the first executable frontier. It\ndoes not reduce a later instruction to implement the spec to that one ticket.\nThe spec links every ticket, and governed implementation retrieves and tracks\nall authorized tickets through their required evidence. Current and retained\nuser authorization determines the endpoint across threads.\n\nCreate or revise exactly one artifact at a time. Begin from the user's ordinary sentence and draft the first useful version yourself. Ask only when an unresolved decision can change the active document; inspect repository facts and use settled conversation context instead of asking the user to repeat discoverable information. Write every settled answer into that Markdown, keep the document concise, and regenerate `\u003Cinitiative-slug>.html`. Remain on the active document until the user clearly approves it.\n\n### Ask by Topic\n\nKeep one Topic per turn. Within that Topic, batch `1–3` mutually related decisions when answering them together reduces back-and-forth. Ask only one when only one meaningful decision remains. Never combine different Topics merely to fill the tool.\n\nCompose the round through the installed helper:\n\n```text\nnode \u003Cskill-dir>\u002Fscripts\u002Ftopic-questions.mjs \\\n  --input \u003Cabsolute-topic-json> \\\n  --format both\n```\n\nThe input contains one `topic`, optional `settledContext`, and `1–3` `decisions`. Each decision has a stable snake-case `id`, a short header, the question, and `2–3` mutually exclusive options. Each option supplies a short label, its impact or trade-off, a concrete example, and exactly one option is marked `recommended: true`. The helper orders the recommendation first, labels it `(Recommended)`, and emits both interfaces from the same source.\n\n- When `request_user_input` is available in the active collaboration mode, call it with the helper's exact `native` payload.\n- When it is unavailable, send the helper's exact `chat` fallback. Do not improvise a different set of options.\n- The client supplies a free-form Other choice for the native interface; the chat fallback explicitly accepts a different option or added nuance.\n- After the reply, persist the selected decisions, rationale, changed assumptions, and remaining unknowns in the active Markdown before moving to the next Topic.\n- If the answer changes the Topic map, add, remove, or reorder later Topics openly. Do not silently preserve stale questions.\n\nThe helper is a deterministic composer, not a source of product judgment. The agent still chooses the relevant Topic, recommendation, trade-offs, and examples from current evidence.\n\nWrite settled plans, not transcripts. Prefer a short summary followed by `3–8` descriptive sections. Keep the evidence needed to understand or approve the decision; move raw logs and exhaustive source notes to referenced evidence files. Never compress away scope, acceptance criteria, risks, decisions, dependencies, or unknowns.\n\nFor a broad initiative with named workstreams, preserve those workstream names as a coverage ledger in every later artifact. Each workstream must be marked `covered`, `unchanged`, `deferred with reason`, or `blocked with evidence`. A Research Report answers residual questions; it never silently replaces the broader Product Grill, story, or convergence scope. The Reader's current page must therefore remain understandable as one phase of the whole initiative.\n\nOn every user reply while `action: review-artifact`, pass the entire message to:\n\n```text\nnode \u003Cskill-dir>\u002Fscripts\u002Ft3-workflow.mjs turn \\\n  --workspace-dir \u003Cabsolute-workspace-dir> \\\n  --repository-path \u003Cabsolute-repository-root> \\\n  --message \u003Centire-user-message> \\\n  --repository-identity \u003Cremote> \\\n  --repository-revision \u003Crevision>\n```\n\nOnly `approval.accepted: true` advances. Ordinary natural-language approval is sufficient for the one active artifact; the user never has to memorize a phrase. A direct affirmation such as `está muy bien`, `me gusta`, `todo correcto`, `perfecto todo aquí`, or `sí` is sufficient. So is a clear no-change statement paired with normal movement language, for example `no tengo cambios, sigue`, `por mí está bien, vamos a la siguiente fase`, `ya quedó, dale`, or `continúa con el siguiente documento`. Feedback revises the same file. Questions, negation, ambiguity, reported approval, combined gates, or a reply that also asks for changes never advance. When affirmation and requested changes coexist, feedback wins: revise the artifact and present its new bytes for approval again.\n\nProduct-convergence prompts may explicitly opt into a reference pack and ask the Technical Grill to infer architecture choices that repository evidence settles. That is prompt-local behavior, not the default for ordinary or complex Working Backwards work.\n\n### Architecture-convergence program\n\nWhen an opted-in product-convergence prompt requests a repository-wide architecture program, add `working_backwards_program: architecture-convergence` to the Technical Grill frontmatter and load the canonical architecture reference pack. The Grill must include the reference pack's complete `Dimension | Current evidence | Decision | Enforcement` matrix for all of these identifiers:\n\n`repository-map`, `module-boundaries`, `dependency-direction`, `file-placement`, `frontend-composition`, `component-design`, `backend-contracts`, `type-contracts`, `testing-strategy`, `documentation`, `performance-security`, `observability`, and `migration-sequencing`.\n\nDo not infer that a repository is understandable merely because its top-level apps have names. Inspect representative paths, imports, public Interfaces, runtime boundaries, tests, docs, generated code, compatibility surfaces, provider adapters, and automation. For components and modules, judge cohesion, responsibility, state ownership, dependency direction, Interface depth, and reuse by proven invariant; never invent a universal line-count limit. For Convex, cover queries, mutations, actions, validators, authorization, indexes, pagination, bounded reads, subscriptions, contention, scheduling, storage, maintained components, and provider adapters.\n\nAgent guardrails, global anti-slop policy, and Release Train design belong to the Development System. Product convergence may verify the installed repository adapter and the product commands it declares, but it must not create product migration workstreams or tickets to redesign those capabilities. Product-specific documentation still explains domain ownership and file\u002Ftest locality; release execution evidence may appear in the final delivery report without making Release Train part of the architecture migration.\n\nAn incomplete matrix fails closed at Technical Grill approval. A row marked `unproven` creates a focused research obligation. Every `change`, `remove`, and unresolved `unproven` row must remain traceable through Product Contract, Technical Contract, and Implementation Map to an executable ticket, an evidence-backed no-change conclusion, or a blocking human gate. Research about one provider or hotspot never replaces the repository-wide architecture program.\n\nAfter an approval, say plainly:\n\n> Terminamos **\u003Cdocumento>**. Siguiente: **\u003Cdocumento siguiente>**.\n\nThen create only the next artifact. Draft it from the approved evidence already available and ask its first high-leverage question only if one remains. Do not dump the remaining process into chat.\n\n## Use the shared technical Reader\n\nThe generated `\u003Cinitiative-slug>.html` is the current-phase entrypoint. Every available prior artifact also has a full Reader page under `reader-history\u002F`; the artifact rail links to those rendered pages, not raw Markdown. Every page states `Fase X de 9` and explains that it is one part of the wider initiative. The format is static, escaped, responsive, offline-capable, and `noindex`. It shows:\n\n- a compact artifact\u002Fphase rail and source links;\n- one continuous `760–900px` technical document;\n- an active `On this page` outline from descriptive headings;\n- restrained title, summary, status, priority, profile, reading time, dates, and repository metadata;\n- the next action and exact human gate in plain language;\n- first-class diagrams, code, tables, decisions, risks, testing, and rollout evidence;\n- `implementationAuthorized: false` until a separate Implement Preview.\n\nOn narrow screens, the document comes first and artifact\u002Foutline navigation becomes secondary controls. The Reader model is plain JSON derived from canonical Markdown and workflow state. Other local development-system surfaces may reuse the same renderer by supplying that model; Working Backwards does not create another presentation system.\n\nUse the Markdown itself to request richer review blocks. They remain canonical\nsource and the shared renderer turns them into safe local visuals:\n\n````markdown\n```mermaid\nflowchart LR\n  A[Idea] --> B{Contrato claro}\n  B -->|Sí| C[Implementación]\n```\n\n```chart\n{\"type\":\"bar\",\"title\":\"Confianza por fase\",\"labels\":[\"Historia\",\"Contrato\",\"Tickets\"],\"values\":[35,72,94]}\n```\n\n```typescript\nexport const outcome = \"visible\";\n```\n````\n\nThe renderer embeds the pinned official Mermaid runtime and supports its diagram families, including flowchart, Gantt\u002Fprogress, sequence, timeline, and architecture, without a home-grown parser. Diagrams provide inline, expanded, and fullscreen reading states; drag\u002Fpan, wheel or trackpad movement, pinch, zoom, fit, reset, copy, and source inspection. Fit preserves a readable Mermaid scale; wide diagrams use the available viewport and pan or scroll horizontally instead of shrinking labels into an unreadable overview. Local bar or line charts use explicit JSON and retain an accessible data table. Semantic Markdown tables, sparse callouts, and fenced code\u002Fdiff blocks preserve filename, language, copy, optional line numbers, highlights, wrapping, and clean horizontal scrolling. Prefer the diagram family that matches the relationship and use a chart only for real comparative or sequential data. Never fabricate metrics.\n\nWhen a final implementation report is requested, make it review-complete rather than a link index. Render the full text or clearly separated Reader sections for every product-convergence prompt or handoff the user explicitly asked to review. Include a **Known issues** table with one row per issue, a disposition of `fixed`, `remaining`, or `authorization-blocked`, and concrete evidence. Links may support those sections but never replace their reviewable content.\n\nKeep the Reader self-contained. Do not add CDN, remote fonts, analytics, hosted rendering, or runtime network dependencies. Never send private planning content outside the operator machine except through the bounded, tokenized, expiring tunnel above when the user has indicated remote review. Do not polish per-feature HTML: improve the shared renderer when the default format needs to change.\n\n## Preserve the three gates\n\nThe first three documents use document approval. Hold exactly three formal definition gates:\n\n- `product-contract` -> `approve-product-contract`\n- `technical-contract` -> `approve-technical-contract`\n- `implementation-map` -> `approve-implementation-map`\n\nThe definition helper persists canonical, hash-bound receipts privately under\n`~\u002F.development-system\u002Fprivate\u002Fworking-backwards\u002F\u003Cworkflow-id>\u002F`, binding the\nabsolute root, Git revision and first executable slice. Its legacy Implement\nPreview operation still accepts only that exact first slice and rejects stale\nor concurrent state. That legacy operation is not the execution endpoint for a\nwhole-spec request. A later authorized whole-feature implementation creates a\ngovernance run containing every linked ticket, retains the definition receipts\nas sources and revalidates current repository facts before planning. Drift does\nnot silently discard tickets or replace the user's retained authorization.\n\nAt `create-private-handoff`, write the final handoff only to the helper's `privateHandoffPath`. Populate its required binding frontmatter from `workflowId`, `gateReceiptPath`, repository evidence, all three receipt hashes, the implementation-map hash, and the exact first slice. Run `render` again. Only `action: handoff-ready` proves the handoff matches the approved planning state.\n\n## Stop before implementation\n\nDefault to Standard. Use Quick only for settled, narrow, reversible behavior on one surface. Use Complex for invariants, authorization, sensitive data, destructive behavior, migration\u002Fbackfill, paid activation, uncertain providers, multiple repositories, or difficult rollback.\n\nThis workflow defines work; it does not execute it. Do not implement, publish tracker tickets, commit, push, open a PR, merge, release, deploy, spend money, or perform destructive operations. When `handoff-ready`, tell the user:\n\n> Working Backwards terminó. El spec enlaza todos los tickets y conserva los criterios de aceptación. En una sesión nueva puedes pedir implementar ese spec completo hasta su endpoint autorizado.\n\n## Governed spec and ticket handoff\n\nUse [the shared definition contract](..\u002Fcoding-orchestration\u002Freferences\u002Fspec-ticket-contract.md)\nfor stable criteria, linked tickets, dependencies, rules and required evidence.\nAstra XHigh authors definition and the implementation plan; a different fresh\nAstra XHigh reviews that plan. Fast researchers supply bounded source facts.\nWhen implementation is authorized, use coding-orchestration and its governed\nJev recipe across the complete feature, including nonvisual behavior. Retrieve\nall tickets linked by the spec; preserve settled decisions across threads.\n",{"name":5,"description":7},"main","9f5b43aadb1a92441d65e9cb74a584cd6fca62bc","artifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards","https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system","https:\u002F\u002Fgithub.com\u002FAO-HyS","https:\u002F\u002Faohys.net\u002F","Software Boutique, we make your projects come true live. ","2026-11-14 03:17:27",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add AO-HyS\u002Fdevelopment-system --skill working-backwards",{"tool":38,"command":39},"bunx","bunx skills add AO-HyS\u002Fdevelopment-system --skill working-backwards",{"tool":41,"command":42},"pnpm","pnpm dlx skills add AO-HyS\u002Fdevelopment-system --skill working-backwards","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002FSKILL.md",true,[46,52,58,62,66,70,74,80,84,88,92,96,100,104,108,112,116,120,124,128,132,136,140],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"agents\u002Fopenai.yaml","REFERENCE","参考文档",230,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fagents\u002Fopenai.yaml",{"path":53,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":56,"sourceHtmlUrl":57},"assets\u002Fatkinson-hyperlegible-next-latin.woff2","ASSET","资产文件",33996,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fassets\u002Fatkinson-hyperlegible-next-latin.woff2",{"path":59,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":60,"sourceHtmlUrl":61},"assets\u002Flicenses\u002Fatkinson-OFL.txt",4599,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fassets\u002Flicenses\u002Fatkinson-OFL.txt",{"path":63,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":64,"sourceHtmlUrl":65},"assets\u002Flicenses\u002Fmermaid-MIT.txt",1089,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fassets\u002Flicenses\u002Fmermaid-MIT.txt",{"path":67,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":68,"sourceHtmlUrl":69},"assets\u002Flicenses\u002Fmonaspace-OFL.txt",4478,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fassets\u002Flicenses\u002Fmonaspace-OFL.txt",{"path":71,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":72,"sourceHtmlUrl":73},"assets\u002Flicenses\u002Fpanzoom-MIT.txt",1059,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fassets\u002Flicenses\u002Fpanzoom-MIT.txt",{"path":75,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":78,"sourceHtmlUrl":79},"assets\u002Fmermaid.min.js","SCRIPT","脚本",3565979,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fassets\u002Fmermaid.min.js",{"path":81,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":82,"sourceHtmlUrl":83},"assets\u002Fmonaspace-neon-latin-400.woff2",44476,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fassets\u002Fmonaspace-neon-latin-400.woff2",{"path":85,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":86,"sourceHtmlUrl":87},"assets\u002Fmonaspace-neon-latin-600.woff2",44632,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fassets\u002Fmonaspace-neon-latin-600.woff2",{"path":89,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":90,"sourceHtmlUrl":91},"assets\u002Fpanzoom.min.js",10125,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fassets\u002Fpanzoom.min.js",{"path":93,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":94,"sourceHtmlUrl":95},"assets\u002Freport-evidence.css",5954,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fassets\u002Freport-evidence.css",{"path":97,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":98,"sourceHtmlUrl":99},"assets\u002Freport.css",4092,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fassets\u002Freport.css",{"path":101,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":102,"sourceHtmlUrl":103},"assets\u002Fvisual-document.css",14980,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fassets\u002Fvisual-document.css",{"path":105,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":106,"sourceHtmlUrl":107},"examples\u002Forchestrator-audit-goals.json",3277,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fexamples\u002Forchestrator-audit-goals.json",{"path":109,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":110,"sourceHtmlUrl":111},"report-reference.md",5776,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Freport-reference.md",{"path":113,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":114,"sourceHtmlUrl":115},"scripts\u002Fpreview-report.mjs",2453,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fscripts\u002Fpreview-report.mjs",{"path":117,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":118,"sourceHtmlUrl":119},"scripts\u002Freader-live.mjs",7308,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fscripts\u002Freader-live.mjs",{"path":121,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":122,"sourceHtmlUrl":123},"scripts\u002Freport-evidence.mjs",6247,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fscripts\u002Freport-evidence.mjs",{"path":125,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":126,"sourceHtmlUrl":127},"scripts\u002Ft3-reader.mjs",78271,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fscripts\u002Ft3-reader.mjs",{"path":129,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":130,"sourceHtmlUrl":131},"scripts\u002Ft3-report.mjs",4188,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fscripts\u002Ft3-report.mjs",{"path":133,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":134,"sourceHtmlUrl":135},"scripts\u002Ft3-workflow.mjs",58717,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fscripts\u002Ft3-workflow.mjs",{"path":137,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":138,"sourceHtmlUrl":139},"scripts\u002Ftopic-questions.mjs",6990,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fscripts\u002Ftopic-questions.mjs",{"path":141,"resourceType":76,"resourceTypeDesc":77,"sizeBytes":142,"sourceHtmlUrl":143},"scripts\u002Fvisual-document.mjs",15499,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.25.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002Fscripts\u002Fvisual-document.mjs",[145,150,157,164,171,178],{"skillNo":146,"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":147,"categories":148,"keywords":149,"publishedAt":19},"01M79EQ4BPYQTX8RWN1Y1J2K1F","https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F2.8.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002FSKILL.md",[],[],{"skillNo":151,"name":152,"nameZh":6,"description":153,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":154,"categories":155,"keywords":156,"publishedAt":19},"01M79EQ2E8P85Y1Y7VNJZMNEJE","flow-implement","Implement a requested software change through verification, corrections, and its authorized delivery endpoint.","https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F2.8.0\u002Fskills\u002Finternal\u002Fflow-implement\u002FSKILL.md",[],[],{"skillNo":158,"name":159,"nameZh":6,"description":160,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":161,"categories":162,"keywords":163,"publishedAt":19},"01M79EQ0PNMKHPJ78F8WMNX06P","development-steward","Run the scheduled, read-only weekly maintenance review for the six explicitly opted-in repositories and prepare one concise private report for Check-in.","https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F2.8.0\u002Fskills\u002Finternal\u002Fdevelopment-steward\u002FSKILL.md",[],[],{"skillNo":165,"name":166,"nameZh":6,"description":167,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":168,"categories":169,"keywords":170,"publishedAt":19},"01M79EPXJ56MX7955GWP1HD4B0","grill-with-docs","Run one recoverable interview for functional, visual, or mixed software grills; reuse equivalent answers and offer the shared HTML questionnaire.","https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F2.7.0\u002Fskills\u002Finternal\u002Fgrill-with-docs\u002FSKILL.md",[],[],{"skillNo":172,"name":173,"nameZh":6,"description":174,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":175,"categories":176,"keywords":177,"publishedAt":19},"01M79EPTFPHYEV40ZG52Z0GVS2","global-agent-guardrails","Install, audit, test, or update the Development System's global destructive-command guard for Codex, Claude Code and T3 Code. Use when configuring agent safety hooks or investigating whether dangerous shell operations are blocked.","https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F2.7.0\u002Fskills\u002Finternal\u002Fglobal-agent-guardrails\u002FSKILL.md",[],[],{"skillNo":179,"name":180,"nameZh":6,"description":181,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":182,"licenseFileUrl":183,"sourceHtmlUrl":184,"categories":185,"keywords":186,"publishedAt":19},"01M79EPR0KPHKAFN7V6CEEDR0Y","impeccable","Design or refine frontend UI using project context and a focused design playbook; also handles visual critique and polish.","Apache-2.0","https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F2.7.0\u002Fskills\u002Fexternal\u002Fimpeccable\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F2.7.0\u002Fskills\u002Fexternal\u002Fimpeccable\u002FSKILL.md",[],[]]