[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7C2BGK0Z7DWWG2ACBX3DQST":3,"skill-same:01M7C2BGK0Z7DWWG2ACBX3DQST":63},{"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":26,"commitSha":27,"skillDirectory":28,"forksCount":29,"repositoryHtmlUrl":30,"repositoryHomepageUrl":6,"ownerHtmlUrl":31,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":32,"install":33,"resources":46},"01M7C2BGK0Z7DWWG2ACBX3DQST","skill-creator",null,"Create a new Agent Skill or revise an existing one (SKILL.md frontmatter, instructions, scripts\u002F, references\u002F, assets\u002F) following the Agent Skills specification, then validate and optionally test it. Use when the user wants to make, write, scaffold, update, refactor, validate, or test a skill, improve a skill's description or triggering, or turn a workflow from the conversation into a reusable skill. Not for system prompts, memory\u002Fnotes files, MCP servers, or hooks.","rockexe0000\u002Fagents-shared-capabilities","rockexe0000","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F19487478?v=4","01M7C2AXQ1KKPDMJ8KAHVXYV9Y","01M7C2AXQQ7DFYTWRK996ENKWC",1,"MIT","https:\u002F\u002Fgithub.com\u002Frockexe0000\u002Fagents-shared-capabilities\u002Fblob\u002Fmain\u002Fskills\u002Fskill-creator\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: skill-creator\ndescription: Create a new Agent Skill or revise an existing one (SKILL.md frontmatter, instructions, scripts\u002F, references\u002F, assets\u002F) following the Agent Skills specification, then validate and optionally test it. Use when the user wants to make, write, scaffold, update, refactor, validate, or test a skill, improve a skill's description or triggering, or turn a workflow from the conversation into a reusable skill. Not for system prompts, memory\u002Fnotes files, MCP servers, or hooks.\nlicense: MIT\nmetadata:\n  short-description: Create, update, and validate Agent Skills\n  based-on: \"anthropics\u002Fskills skill-creator; openai\u002Fcodex sample skill-creator; agentskills.io specification\"\n---\n\n# Skill Creator\n\nCreate or revise skills that help a future agent make better decisions on a specific kind of task, without constraining unrelated work. Applies to any runtime that supports Agent Skills.\n\n## Principles\n\n- **Assume the agent is already capable.** Include only information that changes its decisions or improves its output. Cut generic advice, repeated instructions, speculative edge cases, and examples that don't materially clarify anything.\n- **Preserve the user's intent and scope.** A skill supports the requested task; it should not swap the user's chosen tools, expand the assignment, touch unrelated configuration, or imply permission for extra external actions. Don't promote a single example, past failure, or personal preference into a universal rule. Approval to do a task doesn't widen its scope or permissions; for retrying or externally mutating workflows, define a stopping condition proportional to the risk.\n- **Match specificity to risk.** For open-ended work, describe the outcome and the decision criteria and let the agent choose an approach. For work with a preferred shape, give examples or configurable scripts. Reserve fixed steps, deterministic scripts, and absolute language for cases where deviation causes a concrete problem (correctness, safety, permissions, fragile workflows).\n- **Explain the why.** If you catch yourself writing ALWAYS or NEVER in caps, try explaining the reason instead — a model that understands the purpose handles unanticipated cases better than one following a rule.\n- **Keep discovery cheap and precise.** `name` and `description` are visible before the skill loads. State what it does and when to use it; add exclusions only when they prevent likely misrouting. Avoid exhaustive capability lists and catch-alls.\n- **Disclose progressively.** Keep shared purpose, essential constraints, and routing in `SKILL.md`. Move substantial context-specific material to `references\u002F` and say when to read each file. A simple self-contained skill needs no router or extra files.\n- **No surprises.** A skill's contents should not surprise the user if described to them. Don't write malware, or skills designed for unauthorized access or data exfiltration.\n\n## Anatomy\n\n```text\nskill-name\u002F\n├── SKILL.md        required: YAML frontmatter (name + description) + Markdown instructions\n├── scripts\u002F        optional: deterministic, repeated logic (run it; no need to load it)\n├── references\u002F     optional: docs read only when relevant (schemas, APIs, domain rules, detailed procedures)\n└── assets\u002F         optional: files used in the output (templates, icons, fonts, boilerplate)\n```\n\nCreate only the directories and files the task actually needs. Skip README, install guides, changelogs, and duplicate quick references. Keep references one level deep from `SKILL.md`, linked by relative path; give references over ~300 lines a table of contents or searchable keywords.\n\n## Workflow\n\nScale the process to the request: a complex new skill may need every step; a small edit may need only a focused change and validation.\n\n### 1. Capture intent\n\nIf the conversation already contains the workflow to capture, extract it first: tools used, step order, corrections the user made, input\u002Foutput formats. Establish:\n\n1. What should the skill enable the agent to do?\n2. Which requests or contexts should trigger it?\n3. What does the output look like?\n4. Is testing worthwhile? Objectively checkable outputs (file transforms, data extraction, fixed workflows) benefit from it; subjective ones (writing style, design) usually don't.\n\nAsk only when the missing information matters and can't reasonably be inferred. If the user has already explained the task clearly, proceed.\n\n### 2. Choose the location\n\nRespect a user-specified location. Otherwise use the runtime's documented skills directory — personal (e.g. `~\u002F.claude\u002Fskills\u002F` for Claude Code, `$CODEX_HOME\u002Fskills\u002F` for Codex) or project-level if the skill belongs to one repo.\n\nIf skills in this environment are managed from a source repository and copied or linked into runtimes by a sync tool, edit the source, not the projected copy — the copy gets overwritten on the next sync.\n\nCheck that a skill is the right container: a repeatable procedure with steps, scripts, or templates is a skill; standing behavioral rules, facts, or decision records belong in instruction\u002Fmemory files instead.\n\n### 3. Plan reusable resources\n\nWork backward from realistic requests, and add resources only when the benefit is concrete:\n\n- The same transformation would be rewritten each time → `scripts\u002F`. Prefer dependencies that are reliably present in the target environment (POSIX sh runs almost anywhere); document anything else the script needs.\n- A schema, API, or rule set would be rediscovered each time → `references\u002F\u003Ctopic>.md`.\n- Templates or files that go into the output → `assets\u002F`.\n- Several mutually exclusive modes (e.g. different cloud providers) → one reference per mode; `SKILL.md` holds only the selection criteria so the agent reads just the relevant one.\n\n### 4. Write the frontmatter\n\nMinimal:\n\n```yaml\n---\nname: my-skill\ndescription: \u003CWhat it does>. Use when \u003Ctriggers: phrases users say, contexts, keywords>.\n---\n```\n\nNaming: lowercase letters, digits, single hyphens, ≤64 characters, folder name = `name`. Prefer short, action-oriented names; prefix with a tool or domain when it helps discovery. For field limits, optional fields, and how to write a description that triggers well, read [references\u002Ffrontmatter.md](references\u002Ffrontmatter.md).\n\n### 5. Write the body\n\nWrite what another agent needs to perform the task: the desired outcome, non-obvious context, real constraints, and relevant references or tools. Use the imperative. Give a template when the output format must be exact, and an example or two when style matters. Don't prescribe structure, process, or step counts the task doesn't require. Keep `SKILL.md` under ~500 lines; split detail into `references\u002F` as you approach that.\n\nUse only fictional data in examples and test fixtures. Write credentials as placeholders such as `${API_KEY}` or `\u003CYOUR_TOKEN>`, never real values.\n\nThen reread with fresh eyes: does each section change the agent's decisions? If not, cut it.\n\n### 6. Validate\n\n```sh\nscripts\u002Fvalidate.sh \u003Cpath\u002Fto\u002Fskill>                # spec checks; unknown top-level keys only warn\nscripts\u002Fvalidate.sh --strict-spec \u003Cpath\u002Fto\u002Fskill>  # unknown keys are errors (matches skills-ref validate)\n```\n\nPaths are relative to this skill's directory. It checks frontmatter, naming, length limits, leftover placeholders, relative links, and script permissions. Also run any linter the destination repository uses, and actually execute new or changed scripts.\n\nPassing validation means the format is right, not that the skill makes good decisions. Also confirm the description discriminates, the instructions preserve user intent, and references are discoverable.\n\n### 7. Test (when warranted)\n\nFor skills complex or risky enough that behavioral evidence adds real confidence, run with\u002Fwithout-skill comparisons and triggering tests — read [references\u002Fevaluation.md](references\u002Fevaluation.md). Ordinary creation or small edits don't need this.\n\n### 8. Deliver\n\nSummarize what the skill does, what triggers it, and the validation results. Installing or enabling the skill for a runtime, and granting permission for the commands it runs, are separate steps — don't change those configurations unless the user asks.\n\n## Updating an existing skill\n\n- Read the whole skill first, including callers of scripts and references; confirm a resource is unused before removing it.\n- Keep `name`, the folder name, and existing frontmatter fields; change only what's needed.\n- Make narrow corrections based on real usage or observed failures rather than accumulating universal rules from single cases.\n- For a skill copied from an upstream source, keep edits minimal and note where and why it diverges from upstream.\n","# Skill Creator\n\nCreate or revise skills that help a future agent make better decisions on a specific kind of task, without constraining unrelated work. Applies to any runtime that supports Agent Skills.\n\n## Principles\n\n- **Assume the agent is already capable.** Include only information that changes its decisions or improves its output. Cut generic advice, repeated instructions, speculative edge cases, and examples that don't materially clarify anything.\n- **Preserve the user's intent and scope.** A skill supports the requested task; it should not swap the user's chosen tools, expand the assignment, touch unrelated configuration, or imply permission for extra external actions. Don't promote a single example, past failure, or personal preference into a universal rule. Approval to do a task doesn't widen its scope or permissions; for retrying or externally mutating workflows, define a stopping condition proportional to the risk.\n- **Match specificity to risk.** For open-ended work, describe the outcome and the decision criteria and let the agent choose an approach. For work with a preferred shape, give examples or configurable scripts. Reserve fixed steps, deterministic scripts, and absolute language for cases where deviation causes a concrete problem (correctness, safety, permissions, fragile workflows).\n- **Explain the why.** If you catch yourself writing ALWAYS or NEVER in caps, try explaining the reason instead — a model that understands the purpose handles unanticipated cases better than one following a rule.\n- **Keep discovery cheap and precise.** `name` and `description` are visible before the skill loads. State what it does and when to use it; add exclusions only when they prevent likely misrouting. Avoid exhaustive capability lists and catch-alls.\n- **Disclose progressively.** Keep shared purpose, essential constraints, and routing in `SKILL.md`. Move substantial context-specific material to `references\u002F` and say when to read each file. A simple self-contained skill needs no router or extra files.\n- **No surprises.** A skill's contents should not surprise the user if described to them. Don't write malware, or skills designed for unauthorized access or data exfiltration.\n\n## Anatomy\n\n```text\nskill-name\u002F\n├── SKILL.md        required: YAML frontmatter (name + description) + Markdown instructions\n├── scripts\u002F        optional: deterministic, repeated logic (run it; no need to load it)\n├── references\u002F     optional: docs read only when relevant (schemas, APIs, domain rules, detailed procedures)\n└── assets\u002F         optional: files used in the output (templates, icons, fonts, boilerplate)\n```\n\nCreate only the directories and files the task actually needs. Skip README, install guides, changelogs, and duplicate quick references. Keep references one level deep from `SKILL.md`, linked by relative path; give references over ~300 lines a table of contents or searchable keywords.\n\n## Workflow\n\nScale the process to the request: a complex new skill may need every step; a small edit may need only a focused change and validation.\n\n### 1. Capture intent\n\nIf the conversation already contains the workflow to capture, extract it first: tools used, step order, corrections the user made, input\u002Foutput formats. Establish:\n\n1. What should the skill enable the agent to do?\n2. Which requests or contexts should trigger it?\n3. What does the output look like?\n4. Is testing worthwhile? Objectively checkable outputs (file transforms, data extraction, fixed workflows) benefit from it; subjective ones (writing style, design) usually don't.\n\nAsk only when the missing information matters and can't reasonably be inferred. If the user has already explained the task clearly, proceed.\n\n### 2. Choose the location\n\nRespect a user-specified location. Otherwise use the runtime's documented skills directory — personal (e.g. `~\u002F.claude\u002Fskills\u002F` for Claude Code, `$CODEX_HOME\u002Fskills\u002F` for Codex) or project-level if the skill belongs to one repo.\n\nIf skills in this environment are managed from a source repository and copied or linked into runtimes by a sync tool, edit the source, not the projected copy — the copy gets overwritten on the next sync.\n\nCheck that a skill is the right container: a repeatable procedure with steps, scripts, or templates is a skill; standing behavioral rules, facts, or decision records belong in instruction\u002Fmemory files instead.\n\n### 3. Plan reusable resources\n\nWork backward from realistic requests, and add resources only when the benefit is concrete:\n\n- The same transformation would be rewritten each time → `scripts\u002F`. Prefer dependencies that are reliably present in the target environment (POSIX sh runs almost anywhere); document anything else the script needs.\n- A schema, API, or rule set would be rediscovered each time → `references\u002F\u003Ctopic>.md`.\n- Templates or files that go into the output → `assets\u002F`.\n- Several mutually exclusive modes (e.g. different cloud providers) → one reference per mode; `SKILL.md` holds only the selection criteria so the agent reads just the relevant one.\n\n### 4. Write the frontmatter\n\nMinimal:\n\n```yaml\n---\nname: my-skill\ndescription: \u003CWhat it does>. Use when \u003Ctriggers: phrases users say, contexts, keywords>.\n---\n```\n\nNaming: lowercase letters, digits, single hyphens, ≤64 characters, folder name = `name`. Prefer short, action-oriented names; prefix with a tool or domain when it helps discovery. For field limits, optional fields, and how to write a description that triggers well, read [references\u002Ffrontmatter.md](references\u002Ffrontmatter.md).\n\n### 5. Write the body\n\nWrite what another agent needs to perform the task: the desired outcome, non-obvious context, real constraints, and relevant references or tools. Use the imperative. Give a template when the output format must be exact, and an example or two when style matters. Don't prescribe structure, process, or step counts the task doesn't require. Keep `SKILL.md` under ~500 lines; split detail into `references\u002F` as you approach that.\n\nUse only fictional data in examples and test fixtures. Write credentials as placeholders such as `${API_KEY}` or `\u003CYOUR_TOKEN>`, never real values.\n\nThen reread with fresh eyes: does each section change the agent's decisions? If not, cut it.\n\n### 6. Validate\n\n```sh\nscripts\u002Fvalidate.sh \u003Cpath\u002Fto\u002Fskill>                # spec checks; unknown top-level keys only warn\nscripts\u002Fvalidate.sh --strict-spec \u003Cpath\u002Fto\u002Fskill>  # unknown keys are errors (matches skills-ref validate)\n```\n\nPaths are relative to this skill's directory. It checks frontmatter, naming, length limits, leftover placeholders, relative links, and script permissions. Also run any linter the destination repository uses, and actually execute new or changed scripts.\n\nPassing validation means the format is right, not that the skill makes good decisions. Also confirm the description discriminates, the instructions preserve user intent, and references are discoverable.\n\n### 7. Test (when warranted)\n\nFor skills complex or risky enough that behavioral evidence adds real confidence, run with\u002Fwithout-skill comparisons and triggering tests — read [references\u002Fevaluation.md](references\u002Fevaluation.md). Ordinary creation or small edits don't need this.\n\n### 8. Deliver\n\nSummarize what the skill does, what triggers it, and the validation results. Installing or enabling the skill for a runtime, and granting permission for the commands it runs, are separate steps — don't change those configurations unless the user asks.\n\n## Updating an existing skill\n\n- Read the whole skill first, including callers of scripts and references; confirm a resource is unused before removing it.\n- Keep `name`, the folder name, and existing frontmatter fields; change only what's needed.\n- Make narrow corrections based on real usage or observed failures rather than accumulating universal rules from single cases.\n- For a skill copied from an upstream source, keep edits minimal and note where and why it diverges from upstream.\n",{"name":5,"license":14,"metadata":23,"description":7},{"based-on":24,"short-description":25},"anthropics\u002Fskills skill-creator; openai\u002Fcodex sample skill-creator; agentskills.io specification","Create, update, and validate Agent Skills","main","62aa0268d8bfab607f2b6ad918621d3fe5906c1c","skills\u002Fskill-creator",0,"https:\u002F\u002Fgithub.com\u002Frockexe0000\u002Fagents-shared-capabilities","https:\u002F\u002Fgithub.com\u002Frockexe0000","2026-10-10 16:48:39",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add rockexe0000\u002Fagents-shared-capabilities --skill skill-creator",{"tool":39,"command":40},"bunx","bunx skills add rockexe0000\u002Fagents-shared-capabilities --skill skill-creator",{"tool":42,"command":43},"pnpm","pnpm dlx skills add rockexe0000\u002Fagents-shared-capabilities --skill skill-creator","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Frockexe0000\u002Fagents-shared-capabilities\u002Fblob\u002Fmain\u002Fskills\u002Fskill-creator\u002FSKILL.md",true,[47,53,57],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"references\u002Fevaluation.md","REFERENCE","参考文档",4249,"https:\u002F\u002Fgithub.com\u002Frockexe0000\u002Fagents-shared-capabilities\u002Fblob\u002Fmain\u002Fskills\u002Fskill-creator\u002Freferences\u002Fevaluation.md",{"path":54,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Ffrontmatter.md",2979,"https:\u002F\u002Fgithub.com\u002Frockexe0000\u002Fagents-shared-capabilities\u002Fblob\u002Fmain\u002Fskills\u002Fskill-creator\u002Freferences\u002Ffrontmatter.md",{"path":58,"resourceType":59,"resourceTypeDesc":60,"sizeBytes":61,"sourceHtmlUrl":62},"scripts\u002Fvalidate.sh","SCRIPT","脚本",6366,"https:\u002F\u002Fgithub.com\u002Frockexe0000\u002Fagents-shared-capabilities\u002Fblob\u002Fmain\u002Fskills\u002Fskill-creator\u002Fscripts\u002Fvalidate.sh",[64,71,78,85,92,99],{"skillNo":65,"name":66,"nameZh":6,"description":67,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":68,"categories":69,"keywords":70,"publishedAt":18},"01M7C2BV53E54BXYDPZTR2RGDD","b-skill","fixture skill B.","https:\u002F\u002Fgithub.com\u002Frockexe0000\u002Fagents-shared-capabilities\u002Fblob\u002Fmain\u002Ftools-rs\u002Ftests\u002Ffixtures\u002Fskills-multi\u002Fskills\u002Fb-skill\u002FSKILL.md",[],[],{"skillNo":72,"name":73,"nameZh":6,"description":74,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":75,"categories":76,"keywords":77,"publishedAt":18},"01M7C2BSQCCH587FN1FAG7ERR4","a-skill","fixture skill A (has a nested reference dir to exercise recursive staging).","https:\u002F\u002Fgithub.com\u002Frockexe0000\u002Fagents-shared-capabilities\u002Fblob\u002Fmain\u002Ftools-rs\u002Ftests\u002Ffixtures\u002Fskills-multi\u002Fskills\u002Fa-skill\u002FSKILL.md",[],[],{"skillNo":79,"name":80,"nameZh":6,"description":81,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":82,"categories":83,"keywords":84,"publishedAt":18},"01M7C2BR99AJP8H05VWZEVCSE1","demo-skill","fixture skill that requires a bin tool (drives the requires closure).","https:\u002F\u002Fgithub.com\u002Frockexe0000\u002Fagents-shared-capabilities\u002Fblob\u002Fmain\u002Ftools-rs\u002Ftests\u002Ffixtures\u002Fbin-basic\u002Fskills\u002Fdemo-skill\u002FSKILL.md",[],[],{"skillNo":86,"name":87,"nameZh":6,"description":88,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":89,"categories":90,"keywords":91,"publishedAt":18},"01M7C2BBY60B0QYWRRM9RH84HF","hello-capability","Minimal example skill proving the capability-provisioning pipeline end to end. Use only as a scaffold reference when authoring a real skill; it prints a short confirmation and does nothing else.","https:\u002F\u002Fgithub.com\u002Frockexe0000\u002Fagents-shared-capabilities\u002Fblob\u002Fmain\u002Fskills\u002Fhello-capability\u002FSKILL.md",[],[],{"skillNo":93,"name":94,"nameZh":6,"description":95,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":96,"categories":97,"keywords":98,"publishedAt":18},"01M7C2B4KGTZ5T6RN9X2A5JVBJ","cfdrop","Explain or present anything as a mobile-optimized static website deployed to a temporary Cloudflare account via the cfdrop CLI. Use when the user says \"cfdrop this\", \"explain this by cfdrop\", \"cfdrop mobile optimized\", \"deploy as mobile site\", or asks to turn content (a topic, codebase, issue list, analysis, report) into a browsable mobile web page with a live URL.","https:\u002F\u002Fgithub.com\u002Frockexe0000\u002Fagents-shared-capabilities\u002Fblob\u002Fmain\u002Fskills\u002Fcfdrop\u002FSKILL.md",[],[],{"skillNo":100,"name":101,"nameZh":6,"description":102,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":103,"categories":104,"keywords":105,"publishedAt":18},"01M7C2AZ66D546V1WQM99RXMAF","cfdrop-relay","Push a freshly deployed cfdrop URL to a cfdrop relay so a paired viewer device (e.g. an iPad running the cfdrop viewer app) opens it automatically. Use when the user says \"cfdrop and open on the viewer\", \"push to the relay\", \"show it on the iPad\", or wants a deployed site to appear on a paired screen without sharing the URL manually.","https:\u002F\u002Fgithub.com\u002Frockexe0000\u002Fagents-shared-capabilities\u002Fblob\u002Fmain\u002Fskills\u002Fcfdrop-relay\u002FSKILL.md",[],[]]