[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M79E27H2ZWR7SN5NVXXASBKB":3,"skill-same:01M79E27H2ZWR7SN5NVXXASBKB":80},{"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},"01M79E27H2ZWR7SN5NVXXASBKB","coding-orchestration",null,"Choose delegation, ownership, and verification for non-trivial repository execution with a pinned task contract.","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\u002F1.41.0\u002Fskills\u002Finternal\u002Fcoding-orchestration\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: coding-orchestration\ndescription: Choose delegation, ownership, and verification for non-trivial repository execution with a pinned task contract.\n---\n\n# Coding orchestration\n\nOptimize rápido → bien → barato: reach the complete usable result quickly,\ncorrect it with focused evidence, and keep measured cost and complexity low.\nPreserve behavior, data boundaries and authorization.\n\n## Coordinate the full feature\n\nRead [Jev advisory execution](references\u002Fjev-advisory.md) once for nontrivial work.\nKeep the selected parent. Use fast research, Sol 6.1 High planning, a different\nfresh Sol 6.1 High plan review, bounded writers, independent final review and\nactual behavior evidence. Jev advises at useful routing and correction decisions;\nthe parent records the decision and executes through native tools. Classification\nfailure does not globally block work. Tiny deterministic edits stay direct.\n\n## Resolve the run\n\nThe model selected when the conversation starts remains the orchestrator. It\nchooses workers within the user's provider, quota and capability limits and\nintegrates their results. A roster or native profile is a recommendation, not\nruntime identity or a provider chain. A restricted family stays restricted for\nevery descendant; report a missing browser or vision capability instead of\nsubstituting a different family.\n\nBefore a worker starts, pin one execution contract. It names the objective or\nticket, root and revision, owned paths, expected observable behavior, mode,\nauthorization and endpoint, enumerated checks, evidence receipt, stop\nconditions, and the required completion report. Read\n[the execution contract](references\u002Fexecution-contract.md) for the compact\npacket fields.\n\nUse **exact instructions by default**: give ordered actions, expected\nobservations, checks and stops using the execution contract. Outcome delegation\nis an explicit, reasoned exception for a bounded implementation decision; record\nwhy a recipe would be premature and keep the same acceptance requirements.\n\nThe method is model-agnostic. Preserve the acceptance standard when capability\nor availability changes: reduce packet size and simultaneous decisions, supply\nconcrete examples and observable checkpoints, then correct from the returned\nevidence. A model label never proves capability or acceptance. If an essential\ncapability is unavailable, retain the gap and continue independent work; do not\nsilently substitute a provider or promise equivalent results from every model.\n\n## Execute with useful ownership\n\nThe parent owns decomposition, architecture, integration, conflicts,\nverification, findings and the terminal state. Jev advises this parent; native host tools perform the work. Execute the repository's\nlocal canonical recipe; a skill does not migrate a product's form or other\ndomain architecture to match its own examples.\n\nUse deterministic tools directly and batch independent reads. Keep one writer\nper surface. Parallel workers are allowed only for disjoint paths with no\nshared dependency sequence. Start them from the same pinned contract and\nreceive completion events. Do not send repeated unchanged status requests or\npoll a worker when no new event or observation exists. Reuse a worker for a\nfocused correction while it owns the same surface; verify termination before\ntransferring ownership.\n\nEvery worker packet includes the exact root and branch or revision, owned\nsurface, settled routing, relevant references, constraints, focused checks,\nexpected receipt and known failures when resuming. Workers preserve unrelated\nedits and return changed paths, commands, results and remaining gaps. A clean\ndiff, first implementation or passing lint does not close unverified behavior.\n\nWriter packets: `Done when:` is executable (command plus expected\nobservation); the writer retries up to 3 times and the coordinator reruns it.\n`Outcome:` is optional.\n\n## Automated tests and evidence\n\nNo automated tests anywhere: do not create, run or restore them; delete them\nwhen found (`check-no-tests` and the guard enforce this). Every task includes\nreal verification without being asked: computer use, browser, and the\nrepository's verification CLI and feature map. Report passed \u002F failed \u002F not\nreached with evidence.\n\n\n## UI acceptance and evidence\n\nFor visible work, resolve fixture identity, role, organization and useful\nentities once before the main pass. Give a neutral browser executor exact\naction inputs and the values to record; the parent retains the acceptance\nrubric and judges the resulting values. Use the authorized browser mechanism\nand one driver per session.\n\nWhen visual quality matters, complete Impeccable and independent capable\ncritique first, resolve material findings on this candidate, then call\nevidence-capture for final media. A screenshot or copied file proves neither\nlive harness influence nor product acceptance. Backend-only work does not need\na visual workflow.\n\n## Finish\n\nContinue through implementation, focused checks, corrections and the\nauthorized endpoint. Separate source, local runtime, PR, Preview, production\nand acceptance. Report the candidate, actual observed model\u002Ftool identity,\nchanged paths, checks, evidence references and remaining gaps. If blocked,\nname the concrete dependency and finish independent work; a timeout or missing\nreceipt cannot become a passing result.\n\n## Close every task\n\nEvery task that changed files ends with a report:\n\n1. Write the concise packet (`flow-implement\u002Freferences\u002Fcompletion-report.md`).\n   Its Markdown opens with the four sections the command requires, in order:\n   Qué se hizo, Hallazgos, Qué sigue, Detalle (English: What was done,\n   Findings, What's next, Detail). Real verification (pasó \u002F falló \u002F no se\n   alcanzó, con evidencia) goes in Qué se hizo; a mistake seen twice goes in\n   Hallazgos as a proposed hard rule at the highest level (código >\n   lint\u002FCI\u002Fguard > regla\u002Fskill). No tables unless a real comparison needs one.\n2. Render it with `development-system document --input \u003Cpacket> --json`.\n3. Serve it with working-backwards `reader-live.mjs --tunnel` from a\n   per-report copy directory.\n4. Put the URL in the final answer.\n\nA Stop hook asks once when files changed and no report was produced. If a\nreport is truly not useful, say why in one line. When the user asks for the\nanswer in the chat only, produce no document and no tunnel.\n\nFor host-specific CLI dispatch, read [host-dispatch](references\u002Fhost-dispatch.md).\nFor interruption or ownership recovery, read\n[execution-continuity](references\u002Fexecution-continuity.md). T3 packet details\nare in [worker-reference](worker-reference.md).\n\nFor nontrivial delivery and benchmark results, default to a readable HTML report with passed\u002Ffailed\u002Fnot-reached behavior, timing, evidence and limitations. Markdown is supplementary. Publish sanitized report assets through the user-authorized temporary tunnel and state its actual lifetime conditions; keep secrets and raw transcripts outside the served directory. A post-deadline report cannot convert a failed timed run into success.\n\n## Authorized Claude Code account recovery (1.39.0)\n\nFor Claude Code review and computer use, the operator explicitly authorizes the\nlocal Codex credential-home sequence followed by a fresh native Claude role.\nRun codex-review once; it discovers existing profiles without copying credentials\nand records at most two sequential profile attempts. Read receipt.json as well as findings.\nWhen status is fallback_required and fallback.eligible is true, immediately\ndispatch reviewer (plan, diff or security), visual-reviewer (image review) or\nbrowser-qa (computer use) through the native Agent tool with the original packet.\nCopy fallback.prompt verbatim from the receipt; it contains the original packet\nand Codex fallback:, Codex fallback receipt: and Codex fallback packet: lines.\nThe guard verifies the full dispatched packet, hash, root, mode and Task-Id and\nrejects resume so the native agent stays fresh. Computer-use fallback receipts\nare admitted once only; a consumed receipt requires reconciliation, not replay.\nDo not let a failed Codex attempt silently omit the independent review or absorb\nit into the coordinator. Preserve the endpoint and actual capability requirements.\nCancellation, safety refusal and uncertain computer-use effects are not fallback\npermission: reconcile observed state before more actions. If the native role lacks\nvision or an authorized browser, retain the gap and continue independent work.\nReport the provider change and actual or unknown identity. This explicit Claude\nexception does not change restricted Codex-only tasks or Headroom's same-account\nper-invocation contract. HOME installation does not prove live host loading.\n\n## Generated host profile\n\nUse the installed host instructions generated from config\u002Fagent-roster.json for role\u002Fmodel selection; do not infer a role from older example packets. For T3 cross-provider dispatch and browser capability checks, read coding-orchestration\u002Freferences\u002Ft3-native.md only when needed. Source writers remain on protected routes; scoped product actions are separate authority.\n","# Coding orchestration\n\nOptimize rápido → bien → barato: reach the complete usable result quickly,\ncorrect it with focused evidence, and keep measured cost and complexity low.\nPreserve behavior, data boundaries and authorization.\n\n## Coordinate the full feature\n\nRead [Jev advisory execution](references\u002Fjev-advisory.md) once for nontrivial work.\nKeep the selected parent. Use fast research, Sol 6.1 High planning, a different\nfresh Sol 6.1 High plan review, bounded writers, independent final review and\nactual behavior evidence. Jev advises at useful routing and correction decisions;\nthe parent records the decision and executes through native tools. Classification\nfailure does not globally block work. Tiny deterministic edits stay direct.\n\n## Resolve the run\n\nThe model selected when the conversation starts remains the orchestrator. It\nchooses workers within the user's provider, quota and capability limits and\nintegrates their results. A roster or native profile is a recommendation, not\nruntime identity or a provider chain. A restricted family stays restricted for\nevery descendant; report a missing browser or vision capability instead of\nsubstituting a different family.\n\nBefore a worker starts, pin one execution contract. It names the objective or\nticket, root and revision, owned paths, expected observable behavior, mode,\nauthorization and endpoint, enumerated checks, evidence receipt, stop\nconditions, and the required completion report. Read\n[the execution contract](references\u002Fexecution-contract.md) for the compact\npacket fields.\n\nUse **exact instructions by default**: give ordered actions, expected\nobservations, checks and stops using the execution contract. Outcome delegation\nis an explicit, reasoned exception for a bounded implementation decision; record\nwhy a recipe would be premature and keep the same acceptance requirements.\n\nThe method is model-agnostic. Preserve the acceptance standard when capability\nor availability changes: reduce packet size and simultaneous decisions, supply\nconcrete examples and observable checkpoints, then correct from the returned\nevidence. A model label never proves capability or acceptance. If an essential\ncapability is unavailable, retain the gap and continue independent work; do not\nsilently substitute a provider or promise equivalent results from every model.\n\n## Execute with useful ownership\n\nThe parent owns decomposition, architecture, integration, conflicts,\nverification, findings and the terminal state. Jev advises this parent; native host tools perform the work. Execute the repository's\nlocal canonical recipe; a skill does not migrate a product's form or other\ndomain architecture to match its own examples.\n\nUse deterministic tools directly and batch independent reads. Keep one writer\nper surface. Parallel workers are allowed only for disjoint paths with no\nshared dependency sequence. Start them from the same pinned contract and\nreceive completion events. Do not send repeated unchanged status requests or\npoll a worker when no new event or observation exists. Reuse a worker for a\nfocused correction while it owns the same surface; verify termination before\ntransferring ownership.\n\nEvery worker packet includes the exact root and branch or revision, owned\nsurface, settled routing, relevant references, constraints, focused checks,\nexpected receipt and known failures when resuming. Workers preserve unrelated\nedits and return changed paths, commands, results and remaining gaps. A clean\ndiff, first implementation or passing lint does not close unverified behavior.\n\nWriter packets: `Done when:` is executable (command plus expected\nobservation); the writer retries up to 3 times and the coordinator reruns it.\n`Outcome:` is optional.\n\n## Automated tests and evidence\n\nNo automated tests anywhere: do not create, run or restore them; delete them\nwhen found (`check-no-tests` and the guard enforce this). Every task includes\nreal verification without being asked: computer use, browser, and the\nrepository's verification CLI and feature map. Report passed \u002F failed \u002F not\nreached with evidence.\n\n\n## UI acceptance and evidence\n\nFor visible work, resolve fixture identity, role, organization and useful\nentities once before the main pass. Give a neutral browser executor exact\naction inputs and the values to record; the parent retains the acceptance\nrubric and judges the resulting values. Use the authorized browser mechanism\nand one driver per session.\n\nWhen visual quality matters, complete Impeccable and independent capable\ncritique first, resolve material findings on this candidate, then call\nevidence-capture for final media. A screenshot or copied file proves neither\nlive harness influence nor product acceptance. Backend-only work does not need\na visual workflow.\n\n## Finish\n\nContinue through implementation, focused checks, corrections and the\nauthorized endpoint. Separate source, local runtime, PR, Preview, production\nand acceptance. Report the candidate, actual observed model\u002Ftool identity,\nchanged paths, checks, evidence references and remaining gaps. If blocked,\nname the concrete dependency and finish independent work; a timeout or missing\nreceipt cannot become a passing result.\n\n## Close every task\n\nEvery task that changed files ends with a report:\n\n1. Write the concise packet (`flow-implement\u002Freferences\u002Fcompletion-report.md`).\n   Its Markdown opens with the four sections the command requires, in order:\n   Qué se hizo, Hallazgos, Qué sigue, Detalle (English: What was done,\n   Findings, What's next, Detail). Real verification (pasó \u002F falló \u002F no se\n   alcanzó, con evidencia) goes in Qué se hizo; a mistake seen twice goes in\n   Hallazgos as a proposed hard rule at the highest level (código >\n   lint\u002FCI\u002Fguard > regla\u002Fskill). No tables unless a real comparison needs one.\n2. Render it with `development-system document --input \u003Cpacket> --json`.\n3. Serve it with working-backwards `reader-live.mjs --tunnel` from a\n   per-report copy directory.\n4. Put the URL in the final answer.\n\nA Stop hook asks once when files changed and no report was produced. If a\nreport is truly not useful, say why in one line. When the user asks for the\nanswer in the chat only, produce no document and no tunnel.\n\nFor host-specific CLI dispatch, read [host-dispatch](references\u002Fhost-dispatch.md).\nFor interruption or ownership recovery, read\n[execution-continuity](references\u002Fexecution-continuity.md). T3 packet details\nare in [worker-reference](worker-reference.md).\n\nFor nontrivial delivery and benchmark results, default to a readable HTML report with passed\u002Ffailed\u002Fnot-reached behavior, timing, evidence and limitations. Markdown is supplementary. Publish sanitized report assets through the user-authorized temporary tunnel and state its actual lifetime conditions; keep secrets and raw transcripts outside the served directory. A post-deadline report cannot convert a failed timed run into success.\n\n## Authorized Claude Code account recovery (1.39.0)\n\nFor Claude Code review and computer use, the operator explicitly authorizes the\nlocal Codex credential-home sequence followed by a fresh native Claude role.\nRun codex-review once; it discovers existing profiles without copying credentials\nand records at most two sequential profile attempts. Read receipt.json as well as findings.\nWhen status is fallback_required and fallback.eligible is true, immediately\ndispatch reviewer (plan, diff or security), visual-reviewer (image review) or\nbrowser-qa (computer use) through the native Agent tool with the original packet.\nCopy fallback.prompt verbatim from the receipt; it contains the original packet\nand Codex fallback:, Codex fallback receipt: and Codex fallback packet: lines.\nThe guard verifies the full dispatched packet, hash, root, mode and Task-Id and\nrejects resume so the native agent stays fresh. Computer-use fallback receipts\nare admitted once only; a consumed receipt requires reconciliation, not replay.\nDo not let a failed Codex attempt silently omit the independent review or absorb\nit into the coordinator. Preserve the endpoint and actual capability requirements.\nCancellation, safety refusal and uncertain computer-use effects are not fallback\npermission: reconcile observed state before more actions. If the native role lacks\nvision or an authorized browser, retain the gap and continue independent work.\nReport the provider change and actual or unknown identity. This explicit Claude\nexception does not change restricted Codex-only tasks or Headroom's same-account\nper-invocation contract. HOME installation does not prove live host loading.\n\n## Generated host profile\n\nUse the installed host instructions generated from config\u002Fagent-roster.json for role\u002Fmodel selection; do not infer a role from older example packets. For T3 cross-provider dispatch and browser capability checks, read coding-orchestration\u002Freferences\u002Ft3-native.md only when needed. Source writers remain on protected routes; scoped product actions are separate authority.\n",{"name":5,"description":7},"main","9f5b43aadb1a92441d65e9cb74a584cd6fca62bc","artifacts\u002F1.41.0\u002Fskills\u002Finternal\u002Fcoding-orchestration","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 coding-orchestration",{"tool":38,"command":39},"bunx","bunx skills add AO-HyS\u002Fdevelopment-system --skill coding-orchestration",{"tool":41,"command":42},"pnpm","pnpm dlx skills add AO-HyS\u002Fdevelopment-system --skill coding-orchestration","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.41.0\u002Fskills\u002Finternal\u002Fcoding-orchestration\u002FSKILL.md",true,[46,52,56,60,64,68,72,76],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Fexecution-continuity.md","REFERENCE","参考文档",3056,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.41.0\u002Fskills\u002Finternal\u002Fcoding-orchestration\u002Freferences\u002Fexecution-continuity.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Fexecution-contract.md",3486,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.41.0\u002Fskills\u002Finternal\u002Fcoding-orchestration\u002Freferences\u002Fexecution-contract.md",{"path":57,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":58,"sourceHtmlUrl":59},"references\u002Fhost-dispatch.md",2783,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.41.0\u002Fskills\u002Finternal\u002Fcoding-orchestration\u002Freferences\u002Fhost-dispatch.md",{"path":61,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":62,"sourceHtmlUrl":63},"references\u002Fjev-advisory.md",3030,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.41.0\u002Fskills\u002Finternal\u002Fcoding-orchestration\u002Freferences\u002Fjev-advisory.md",{"path":65,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":66,"sourceHtmlUrl":67},"references\u002Fjev-governance.md",428,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.41.0\u002Fskills\u002Finternal\u002Fcoding-orchestration\u002Freferences\u002Fjev-governance.md",{"path":69,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":70,"sourceHtmlUrl":71},"references\u002Fspec-ticket-contract.md",3288,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.41.0\u002Fskills\u002Finternal\u002Fcoding-orchestration\u002Freferences\u002Fspec-ticket-contract.md",{"path":73,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":74,"sourceHtmlUrl":75},"references\u002Ft3-native.md",1829,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.41.0\u002Fskills\u002Finternal\u002Fcoding-orchestration\u002Freferences\u002Ft3-native.md",{"path":77,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":78,"sourceHtmlUrl":79},"worker-reference.md",885,"https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F1.41.0\u002Fskills\u002Finternal\u002Fcoding-orchestration\u002Fworker-reference.md",[81,88,95,102,109,116],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":19},"01M79EQ4BPYQTX8RWN1Y1J2K1F","working-backwards","Define a software feature through PRFAQ and staged documents when the user requests working backwards before coding.","https:\u002F\u002Fgithub.com\u002FAO-HyS\u002Fdevelopment-system\u002Fblob\u002Fmain\u002Fartifacts\u002F2.8.0\u002Fskills\u002Finternal\u002Fworking-backwards\u002FSKILL.md",[],[],{"skillNo":89,"name":90,"nameZh":6,"description":91,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":92,"categories":93,"keywords":94,"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":96,"name":97,"nameZh":6,"description":98,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":99,"categories":100,"keywords":101,"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":103,"name":104,"nameZh":6,"description":105,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":106,"categories":107,"keywords":108,"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":110,"name":111,"nameZh":6,"description":112,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":113,"categories":114,"keywords":115,"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":117,"name":118,"nameZh":6,"description":119,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":120,"licenseFileUrl":121,"sourceHtmlUrl":122,"categories":123,"keywords":124,"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",[],[]]