[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7D2PNFGERY2T2W4GGR3E925":3,"skill-same:01M7D2PNFGERY2T2W4GGR3E925":49},{"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":27,"commitSha":28,"skillDirectory":29,"forksCount":30,"repositoryHtmlUrl":31,"repositoryHomepageUrl":6,"ownerHtmlUrl":32,"ownerBlogUrl":33,"ownerDescription":6,"updatedAt":34,"install":35,"resources":48},"01M7D2PNFGERY2T2W4GGR3E925","compound",null,"Human-started command: it runs only when the human's message begins with `\u002Fcompound`, or when a shipping or autopilot run reaches its lesson operation. On any other request do not load or follow it; tell the human to type that command. Captures one reusable lesson from a real failure, a correction or a surprising review result, as a candidate in the central knowledgebase. Use when verified work produced reasoning that the final code, tests and existing lessons do not already carry. Not for routine runs that simply ended.","Pibomeister\u002Fagent-kit","Pibomeister","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F11633791?v=4","01M7D2NBHEKARMHN9MPV2J80XD","Eduardo Picazo","01M7D2NBJ3Z33SGSYCZ0KBD5DE",0,"MIT","https:\u002F\u002Fgithub.com\u002FPibomeister\u002Fagent-kit\u002Fblob\u002Fmain\u002F.cursor\u002Fskills\u002Fcompound\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: compound\ndescription: \"Human-started command: it runs only when the human's message begins with `\u002Fcompound`, or when a shipping or autopilot run reaches its lesson operation. On any other request do not load or follow it; tell the human to type that command. Captures one reusable lesson from a real failure, a correction or a surprising review result, as a candidate in the central knowledgebase. Use when verified work produced reasoning that the final code, tests and existing lessons do not already carry. Not for routine runs that simply ended.\"\nlicense: MIT\nmetadata:\n  ak_catalog_id: compound\n  ak:\n    mode: manual\n---\n\nCapture a reusable lesson tied to a real failure, correction or surprising review result. No\nmanufactured lesson just because a run ended.\n\n## When to use\n\n- A problem was solved and verified, and the reasoning that solved it is not recoverable from the\n  final code, tests, types or the project's existing lessons.\n- A human corrected the agent's approach, and the correction would apply to future work.\n- A review raised something nobody expected, and the surprise is about the project rather than\n  about one line.\n- An existing lesson turned out to be materially wrong or incomplete in the work just done: update\n  it rather than adding a second one.\n\n## Not for\n\n- A run that went as planned. Completion, effort and diff size do not make a lesson.\n- Auditing, merging or retiring lessons already captured. That is `compound-refresh`, a separate\n  human-started run.\n- Writing anything into the working repository. Lessons are project knowledge and the central\n  knowledgebase owns them (ruling `central-kb-owns-project-artifacts`).\n- Promoting the learning runtime's ledger entries wholesale. A confirmed ledger lesson is evidence\n  for a capture, never a substitute for one (ruling `learning-drafts-not-publishes`).\n\n## Authority\n\nAuthority: `explicit` at the public entrypoint, `model` at the phase operation lesson.capture, and\n`explicit-or-delegated` at `lesson.publish`. A human starts the public entrypoint by typing\n`\u002Fcompound`. A request in prose is not a start, even when it names this skill or the command.\n`super-ship` and `autopilot` reach this skill only through lesson.capture, which drafts a candidate\nfrom evidence already in the run and never publishes. Publishing runs through\n`lesson.publish`: a human's explicit say-so in the session, or a runner-validated grant covering\n`publish-lesson`. Where the host cannot validate that grant, publishing stops for explicit\ninvocation and the candidate stays a draft (ruling `entrypoint-phase-operation-split`).\n\n## Inputs\n\n- The run's evidence: the failure, the correction or the review finding, referenced by artifact id\n  and hash. Absent: `complete` with no lesson, and the report says there was no trigger.\n- The knowledgebase adapter's `readContext`, to find an existing lesson on the same subject.\n  Unavailable: `needs-input`.\n- Optional, under the `learning` profile: `.cursor\u002Fbin\u002Fak learn review report` for the repository's active\n  review patterns and guardrail drafts, and `.cursor\u002Fbin\u002Fak learn memory show` for its working memory and\n  confirmed lessons. A pattern or lesson there is admitted as supporting evidence only where the\n  evidence gate admitted it (protocol [`evidence-gate`](..\u002F..\u002Freferences\u002Fshared\u002Fprotocols\u002Fevidence-gate\u002FPROTOCOL.md)); a `candidate` pattern or a `hypothesis`\n  lesson is context, not support. These are runtime commands, not skills, and the runtime is an\n  optional host adapter; its absence changes nothing below (ruling\n  `learning-runtime-is-host-adapter`).\n\n## Workflow\n\n1. **Check how this run was started**, before any other step and before any tool call. It is started\n   only when the human's message begins with `\u002Fcompound`, when a `super-ship` or `autopilot` run\n   reached lesson.capture, or when `lesson.publish` runs under a validated grant. A request in prose\n   is not a start, even when it names this skill or the command, or asks for this work without\n   naming either. With none of the three, stop before reading the artifacts, checking inputs or\n   answering the task: the only response is to tell the human to type `\u002Fcompound` followed by\n   their request.\n2. **Find the trigger.** Name the failure, correction or surprising review result, and the artifact\n   where it happened. No trigger: stop with no lesson and say so.\n3. **Apply the counterfactual.** If this lesson disappeared, would a future engineer reading the\n   final implementation still be likely to repeat the mistake or redo substantial investigation? If\n   not, write nothing and report why. An explicit invocation asks for the judgment now; it does not\n   lower the bar.\n4. **Look for the lesson already written.** Search the knowledgebase for the subject, and the\n   runtime's ledgers when the profile is installed. A matching lesson that is still right means\n   nothing to capture. A matching lesson made wrong or incomplete by this work is superseded: the\n   new candidate names it in `supersedes`.\n5. **Draft one lesson.** A statement that applies to future work, its trigger bound to the\n   occurrence, the evidence that shows it is real, and where it applies. One lesson per run: a\n   session that produced several gets several runs, because a batched capture blurs which evidence\n   supports which statement. Load\n   [the prose-quality reference pack](..\u002F..\u002Freferences\u002Fshared\u002Freferences\u002Fprose-quality\u002FREFERENCE.md) before wording it.\n6. **Propose it** through the adapter's `proposeLesson` as a `candidate`\n   ([`schemas\u002Flesson.schema.json`](..\u002F..\u002Freferences\u002Fshared\u002Fschemas\u002Flesson.schema.json)).\n7. **Publish only on authority.** At the public entrypoint, publish when the human says so in this\n   session. At lesson.capture, never. At `lesson.publish`, only under the grant.\n\n## Hard gates\n\nGate: no lesson without a trigger. A run that simply ended produces nothing, and the report says\nso (release scenario 23).\n\nGate: a candidate is a draft. Proposing it is not publishing it, and a completed ship is not\npermission to rewrite project knowledge (ruling `learning-drafts-not-publishes`).\n\nGate: one lesson per run.\n\nGate: nothing is written into the working repository, whatever the knowledgebase's availability.\n\n| The thought | Why it is wrong | Do this instead |\n|---|---|---|\n| \"The run went smoothly, but it was long; there must be something worth recording.\" | Effort is not evidence, and a lesson with no trigger is the manufactured output scenario 23 forbids. | Report that nothing qualifies and stop. |\n| \"The ledger says this pattern is active, so the lesson is proven.\" | Active is a count of events, not a statement that holds for future work, and a ledger entry is a draft. | Cite the pattern as evidence and apply the counterfactual to the statement you draft. |\n| \"Ship finished and the human seemed pleased; I'll publish the lesson.\" | Publishing needs explicit authorization or a `publish-lesson` grant; approval of the ship is neither. | Propose the candidate and report that it awaits publication. |\n| \"The knowledgebase is down; I'll save the lesson under docs in the repo for now.\" | Project knowledge written into the repository is the failure the central knowledgebase exists to prevent. | Stop with `needs-input` and hand back the drafted text. |\n| \"These three things all came out of one session; one lesson can cover them.\" | A batched lesson loses which evidence supports which claim, and it will be refreshed as one unit. | Capture one. Recommend a separate run for each of the others. |\n\n## Outputs\n\n- `lesson` ([`schemas\u002Flesson.schema.json`](..\u002F..\u002Freferences\u002Fshared\u002Fschemas\u002Flesson.schema.json)), status `candidate`, through `proposeLesson`, with its\n  trigger, evidence, where it applies and anything it supersedes.\n- On explicit authorization or under the grant, the same lesson published.\n- A report: the trigger, the counterfactual answer, the lesson or the reason there is none, and\n  whether it was published or awaits publication.\n\n## Side effects\n\n`artifact-write`, `scratch-write`, `kb-draft`, `kb-publish`. No `workspace-write`: this skill\nnever edits the repository it learned from.\n\n`kb-publish` is a remote side effect. Its idempotency key derives from the run, the operation, the\nlesson's identity and its artifact hash, and the read-back is the lesson as the knowledgebase\nreturns it ([`adapters\u002Frunner-contract\u002FCONTRACT.md`](..\u002F..\u002Freferences\u002Fshared\u002Fadapters\u002Frunner-contract\u002FCONTRACT.md)). A publish whose read-back cannot be performed\nis `failed`, never complete.\n\n## Stop conditions\n\n- `complete` — one candidate proposed, published where authorized; or no trigger, or the\n  counterfactual failed, and the report says which.\n- `needs-input` — the run was started by neither the typed command, a phase operation nor a\n  validated grant, which returns the command to type and nothing else; the knowledgebase adapter\n  is unavailable; or publishing was asked for without authority the host can validate.\n- `failed` — a write's read-back could not be performed.\n\n## Limits\n\n- Lessons per run: 1 (gate).\n- Ledger evidence: only what passed the evidence gate counts as support (gate).\n- Lesson statement length: short enough to apply without rereading the session (guidance).\n","Capture a reusable lesson tied to a real failure, correction or surprising review result. No\nmanufactured lesson just because a run ended.\n\n## When to use\n\n- A problem was solved and verified, and the reasoning that solved it is not recoverable from the\n  final code, tests, types or the project's existing lessons.\n- A human corrected the agent's approach, and the correction would apply to future work.\n- A review raised something nobody expected, and the surprise is about the project rather than\n  about one line.\n- An existing lesson turned out to be materially wrong or incomplete in the work just done: update\n  it rather than adding a second one.\n\n## Not for\n\n- A run that went as planned. Completion, effort and diff size do not make a lesson.\n- Auditing, merging or retiring lessons already captured. That is `compound-refresh`, a separate\n  human-started run.\n- Writing anything into the working repository. Lessons are project knowledge and the central\n  knowledgebase owns them (ruling `central-kb-owns-project-artifacts`).\n- Promoting the learning runtime's ledger entries wholesale. A confirmed ledger lesson is evidence\n  for a capture, never a substitute for one (ruling `learning-drafts-not-publishes`).\n\n## Authority\n\nAuthority: `explicit` at the public entrypoint, `model` at the phase operation lesson.capture, and\n`explicit-or-delegated` at `lesson.publish`. A human starts the public entrypoint by typing\n`\u002Fcompound`. A request in prose is not a start, even when it names this skill or the command.\n`super-ship` and `autopilot` reach this skill only through lesson.capture, which drafts a candidate\nfrom evidence already in the run and never publishes. Publishing runs through\n`lesson.publish`: a human's explicit say-so in the session, or a runner-validated grant covering\n`publish-lesson`. Where the host cannot validate that grant, publishing stops for explicit\ninvocation and the candidate stays a draft (ruling `entrypoint-phase-operation-split`).\n\n## Inputs\n\n- The run's evidence: the failure, the correction or the review finding, referenced by artifact id\n  and hash. Absent: `complete` with no lesson, and the report says there was no trigger.\n- The knowledgebase adapter's `readContext`, to find an existing lesson on the same subject.\n  Unavailable: `needs-input`.\n- Optional, under the `learning` profile: `.cursor\u002Fbin\u002Fak learn review report` for the repository's active\n  review patterns and guardrail drafts, and `.cursor\u002Fbin\u002Fak learn memory show` for its working memory and\n  confirmed lessons. A pattern or lesson there is admitted as supporting evidence only where the\n  evidence gate admitted it (protocol [`evidence-gate`](..\u002F..\u002Freferences\u002Fshared\u002Fprotocols\u002Fevidence-gate\u002FPROTOCOL.md)); a `candidate` pattern or a `hypothesis`\n  lesson is context, not support. These are runtime commands, not skills, and the runtime is an\n  optional host adapter; its absence changes nothing below (ruling\n  `learning-runtime-is-host-adapter`).\n\n## Workflow\n\n1. **Check how this run was started**, before any other step and before any tool call. It is started\n   only when the human's message begins with `\u002Fcompound`, when a `super-ship` or `autopilot` run\n   reached lesson.capture, or when `lesson.publish` runs under a validated grant. A request in prose\n   is not a start, even when it names this skill or the command, or asks for this work without\n   naming either. With none of the three, stop before reading the artifacts, checking inputs or\n   answering the task: the only response is to tell the human to type `\u002Fcompound` followed by\n   their request.\n2. **Find the trigger.** Name the failure, correction or surprising review result, and the artifact\n   where it happened. No trigger: stop with no lesson and say so.\n3. **Apply the counterfactual.** If this lesson disappeared, would a future engineer reading the\n   final implementation still be likely to repeat the mistake or redo substantial investigation? If\n   not, write nothing and report why. An explicit invocation asks for the judgment now; it does not\n   lower the bar.\n4. **Look for the lesson already written.** Search the knowledgebase for the subject, and the\n   runtime's ledgers when the profile is installed. A matching lesson that is still right means\n   nothing to capture. A matching lesson made wrong or incomplete by this work is superseded: the\n   new candidate names it in `supersedes`.\n5. **Draft one lesson.** A statement that applies to future work, its trigger bound to the\n   occurrence, the evidence that shows it is real, and where it applies. One lesson per run: a\n   session that produced several gets several runs, because a batched capture blurs which evidence\n   supports which statement. Load\n   [the prose-quality reference pack](..\u002F..\u002Freferences\u002Fshared\u002Freferences\u002Fprose-quality\u002FREFERENCE.md) before wording it.\n6. **Propose it** through the adapter's `proposeLesson` as a `candidate`\n   ([`schemas\u002Flesson.schema.json`](..\u002F..\u002Freferences\u002Fshared\u002Fschemas\u002Flesson.schema.json)).\n7. **Publish only on authority.** At the public entrypoint, publish when the human says so in this\n   session. At lesson.capture, never. At `lesson.publish`, only under the grant.\n\n## Hard gates\n\nGate: no lesson without a trigger. A run that simply ended produces nothing, and the report says\nso (release scenario 23).\n\nGate: a candidate is a draft. Proposing it is not publishing it, and a completed ship is not\npermission to rewrite project knowledge (ruling `learning-drafts-not-publishes`).\n\nGate: one lesson per run.\n\nGate: nothing is written into the working repository, whatever the knowledgebase's availability.\n\n| The thought | Why it is wrong | Do this instead |\n|---|---|---|\n| \"The run went smoothly, but it was long; there must be something worth recording.\" | Effort is not evidence, and a lesson with no trigger is the manufactured output scenario 23 forbids. | Report that nothing qualifies and stop. |\n| \"The ledger says this pattern is active, so the lesson is proven.\" | Active is a count of events, not a statement that holds for future work, and a ledger entry is a draft. | Cite the pattern as evidence and apply the counterfactual to the statement you draft. |\n| \"Ship finished and the human seemed pleased; I'll publish the lesson.\" | Publishing needs explicit authorization or a `publish-lesson` grant; approval of the ship is neither. | Propose the candidate and report that it awaits publication. |\n| \"The knowledgebase is down; I'll save the lesson under docs in the repo for now.\" | Project knowledge written into the repository is the failure the central knowledgebase exists to prevent. | Stop with `needs-input` and hand back the drafted text. |\n| \"These three things all came out of one session; one lesson can cover them.\" | A batched lesson loses which evidence supports which claim, and it will be refreshed as one unit. | Capture one. Recommend a separate run for each of the others. |\n\n## Outputs\n\n- `lesson` ([`schemas\u002Flesson.schema.json`](..\u002F..\u002Freferences\u002Fshared\u002Fschemas\u002Flesson.schema.json)), status `candidate`, through `proposeLesson`, with its\n  trigger, evidence, where it applies and anything it supersedes.\n- On explicit authorization or under the grant, the same lesson published.\n- A report: the trigger, the counterfactual answer, the lesson or the reason there is none, and\n  whether it was published or awaits publication.\n\n## Side effects\n\n`artifact-write`, `scratch-write`, `kb-draft`, `kb-publish`. No `workspace-write`: this skill\nnever edits the repository it learned from.\n\n`kb-publish` is a remote side effect. Its idempotency key derives from the run, the operation, the\nlesson's identity and its artifact hash, and the read-back is the lesson as the knowledgebase\nreturns it ([`adapters\u002Frunner-contract\u002FCONTRACT.md`](..\u002F..\u002Freferences\u002Fshared\u002Fadapters\u002Frunner-contract\u002FCONTRACT.md)). A publish whose read-back cannot be performed\nis `failed`, never complete.\n\n## Stop conditions\n\n- `complete` — one candidate proposed, published where authorized; or no trigger, or the\n  counterfactual failed, and the report says which.\n- `needs-input` — the run was started by neither the typed command, a phase operation nor a\n  validated grant, which returns the command to type and nothing else; the knowledgebase adapter\n  is unavailable; or publishing was asked for without authority the host can validate.\n- `failed` — a write's read-back could not be performed.\n\n## Limits\n\n- Lessons per run: 1 (gate).\n- Ledger evidence: only what passed the evidence gate counts as support (gate).\n- Lesson statement length: short enough to apply without rereading the session (guidance).\n",{"name":5,"license":15,"metadata":24,"description":7},{"ak":25,"ak_catalog_id":5},{"mode":26},"manual","main","e0232821c4abbcd4338d6018c20fb70f1b74336b",".cursor\u002Fskills\u002Fcompound",3,"https:\u002F\u002Fgithub.com\u002FPibomeister\u002Fagent-kit","https:\u002F\u002Fgithub.com\u002FPibomeister","www.eduardopicazoenriquez.com","2026-11-15 00:56:58",{"source":8,"skillName":5,"commands":36,"prompt":46,"zipAvailable":47},[37,40,43],{"tool":38,"command":39},"npx","npx skills add Pibomeister\u002Fagent-kit --skill compound",{"tool":41,"command":42},"bunx","bunx skills add Pibomeister\u002Fagent-kit --skill compound",{"tool":44,"command":45},"pnpm","pnpm dlx skills add Pibomeister\u002Fagent-kit --skill compound","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FPibomeister\u002Fagent-kit\u002Fblob\u002Fmain\u002F.cursor\u002Fskills\u002Fcompound\u002FSKILL.md",true,[],[50,57,62,67,72,77],{"skillNo":51,"name":52,"nameZh":6,"description":53,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":54,"categories":55,"keywords":56,"publishedAt":19},"01M7D3JZP2026RGY6XX1MC3NQQ","alpha","Human-started command: it runs only when the human's message begins with `\u002Fak:alpha`. On any other request do not load or follow it; tell the human to type that command. Runs the alpha workflow when a human asks for it.","https:\u002F\u002Fgithub.com\u002FPibomeister\u002Fagent-kit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fvalid\u002Fminimal\u002Fskills\u002Falpha\u002FSKILL.md",[],[],{"skillNo":58,"name":52,"nameZh":6,"description":53,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":59,"categories":60,"keywords":61,"publishedAt":19},"01M7D3JA593Q3RBYJXKVZ37Y2K","https:\u002F\u002Fgithub.com\u002FPibomeister\u002Fagent-kit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Finvalid\u002F25-article-tier-outside-vocabulary\u002Fskills\u002Falpha\u002FSKILL.md",[],[],{"skillNo":63,"name":52,"nameZh":6,"description":53,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":64,"categories":65,"keywords":66,"publishedAt":19},"01M7D3HQDDAZTAPGWCPGDPNVBQ","https:\u002F\u002Fgithub.com\u002FPibomeister\u002Fagent-kit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Finvalid\u002F24-article-without-triggers\u002Fskills\u002Falpha\u002FSKILL.md",[],[],{"skillNo":68,"name":52,"nameZh":6,"description":53,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":69,"categories":70,"keywords":71,"publishedAt":19},"01M7D3HK074TK3ZWK1ZZNQVPB0","https:\u002F\u002Fgithub.com\u002FPibomeister\u002Fagent-kit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Finvalid\u002F23-enforced-by-without-predicate\u002Fskills\u002Falpha\u002FSKILL.md",[],[],{"skillNo":73,"name":52,"nameZh":6,"description":53,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M7D3HGHQM5C5TBSXBCE2WFHW","https:\u002F\u002Fgithub.com\u002FPibomeister\u002Fagent-kit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Finvalid\u002F22-article-without-source-span\u002Fskills\u002Falpha\u002FSKILL.md",[],[],{"skillNo":78,"name":52,"nameZh":6,"description":53,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M7D3HDV6HMK2ZHRM0P368KNT","https:\u002F\u002Fgithub.com\u002FPibomeister\u002Fagent-kit\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Finvalid\u002F21-fragment-redeclares-id\u002Fskills\u002Falpha\u002FSKILL.md",[],[]]