[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7CZCZGDGNX4A0F1DBHT19Z8":3,"skill-same:01M7CZCZGDGNX4A0F1DBHT19Z8":60},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"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":32,"commitSha":33,"skillDirectory":34,"forksCount":35,"repositoryHtmlUrl":30,"repositoryHomepageUrl":6,"ownerHtmlUrl":36,"ownerBlogUrl":37,"ownerDescription":38,"updatedAt":39,"install":40,"resources":53},"01M7CZCZGDGNX4A0F1DBHT19Z8","padosoft-knowhow-harvest",null,"Use this skill to turn what other repositories have learned into skills — a periodic harvest of their lessons files, rule folders, internal skills, decision records and security docs, deciding for each finding whether it updates an existing skill, becomes a new one, or is deliberately left alone. Also when the user says \"check the repos for new rules\", \"harvest the know-how\", \"what changed since last time\", asks how to keep the catalogue current, wants to know whether something was already considered, or asks what to do with the local rules a skill now duplicates. It covers the work order, the judgement that turns a finding into a rule, the promotion criterion, and the ledger that stops the same material being mined twice. Do not use it to write a single skill from scratch (padosoft-skill-creator) or to audit an existing one.","padosoft\u002Fskills","padosoft","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F10467753?v=4","01M7CZ3FRSM851DK225E3KK2VF","01M7CZ3FSY016E21VS6EABM21V",1,"MIT","https:\u002F\u002Fgithub.com\u002Fpadosoft\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fpadosoft-knowhow-harvest\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: padosoft-knowhow-harvest\ndescription: >-\n  Use this skill to turn what other repositories have learned into skills — a periodic harvest of their\n  lessons files, rule folders, internal skills, decision records and security docs, deciding for each\n  finding whether it updates an existing skill, becomes a new one, or is deliberately left alone. Also when\n  the user says \"check the repos for new rules\", \"harvest the know-how\", \"what changed since last time\",\n  asks how to keep the catalogue current, wants to know whether something was already considered, or asks what to do with the local\n  rules a skill now duplicates. It\n  covers the work order, the judgement that turns a finding into a rule, the promotion criterion, and the\n  ledger that stops the same material being mined twice. Do not use it to write a single skill from scratch\n  (padosoft-skill-creator) or to audit an existing one.\nlicense: MIT\ncompatibility: >-\n  Requires the repository's `harvest.py` (in its `scripts\u002F` folder) and read access to the source repositories. The\n  sources and the ledger live in gitignored files, because a list of the repositories you run is metadata\n  about your business.\nmetadata:\n  version: 0.2.0\n  author: Padosoft\n  summary: Harvest the rule, leave the story, and record what you decided so nobody mines it twice.\n  profiles: core\n  scope: global\n  repository: https:\u002F\u002Fgithub.com\u002Fpadosoft\u002Fskills\n  keywords: harvest, lessons, rules, catalogue, maintenance, promotion, ledger, know-how, periodic\n---\n\n# Know-how harvest\n\nEvery repository you run accumulates hard-won rules — in a lessons file, a rules folder, decision records,\nan internal skill. They stay there, benefiting one project. This is the loop that moves them into the shared\ncatalogue **without** re-reading everything every time, and without losing the record of what was decided.\n\n**The rule: harvest the rule, leave the story, and write down what you decided — including the rejections,\nbecause an undocumented rejection is re-mined every six months.**\n\n---\n\n## 0. Setup, once\n\n```bash\ncp .harvest-sources.example.json .harvest-sources.json   # then fill in your paths\n```\n\nBoth that file and `.harvest-ledger.json` are gitignored on purpose: **a list of the repositories you run is\nmetadata about your business**, and this repository is public. If more than one person harvests, keep them\nin a private companion repository or a secret rather than committing them here.\n\n## 1. The work order\n\n```bash\nmake harvest                 # or: python3 scripts\u002Fharvest.py scan\npython3 scripts\u002Fharvest.py scan --id monolith       # one source\npython3 scripts\u002Fharvest.py scan --json              # to drive another tool\n```\n\nIt prints four groups, and the current catalogue to match against:\n\n| Group | What it means |\n|---|---|\n| **New** | never considered — the bulk on a first run, a trickle afterwards |\n| **Changed** | reconsider; the work order says what you decided last time |\n| **Gone** | the source file is gone — check whether a skill still rests on it |\n| **Unreadable** | too large to read in one pass. Split it or exclude it **deliberately** |\n\nThe problems section is part of the output, not a warning: a source path that no longer resolves is a\nfinding. A harvest that silently skips a repository reports a clean catalogue — see\n**`padosoft-evidence-boundaries`**.\n\n**The scan never writes to the ledger.** A session that dies half way must not leave everything marked as\nprocessed.\n\n## 2. Read for the rule, not for the change\n\nFor each finding, open the file — for a changed one, the **added** material is what matters. Then ask, in\nthis order:\n\n1. **Is it a rule, or is it a note?** A rule states what goes wrong and what prevents it. A dated progress\n   entry, a local path, a vendor version, a review comment is a note: it was true for that repository on\n   that day and carries nothing.\n2. **Would the agent get it wrong without it?** If the model already does the right thing unprompted, the\n   rule adds context and not quality. That is the filter that keeps the catalogue small enough to be read.\n3. **Is it project-specific, or a class of defect?** *A named table must be registered in the resolver* is\n   specific; *a resolver that returns its input unchanged makes an incomplete registration\n   indistinguishable from a no-op* is the class. Harvest the class.\n\nMost findings fail one of the three. That is normal, and saying so explicitly is the point of the ledger.\n\n## 3. The four outcomes\n\n| Outcome | When | What you do |\n|---|---|---|\n| **Extend an existing skill** | the class already has a home | add the section; bump the version; say which skill in the ledger |\n| **Create a new skill** | a coherent domain nothing covers | **`padosoft-skill-creator`**, then register it in a package |\n| **Promote to global** | see §4 | move to `core`, state the boundaries in the description |\n| **Reject** | a note, already covered, too specific, or the agent gets it right anyway | record it **with the reason** |\n\nDefault to extending. A new skill that overlaps an existing one makes both trigger worse, and the\ndescription is what pays the price.\n\n## 4. Promotion is decided by independent convergence\n\nA rule belongs in `core` when **stacks that share no line of code arrived at it on their own**. Not when it\nfeels important, not when it appears in the biggest repository — when two or three unrelated sources\nproduce it independently.\n\n```bash\npython3 scripts\u002Fharvest.py status     # shows how many sources feed each skill\n```\n\nThat count is the evidence. There is no cap: if four sources converge on a fifth rule, it gets promoted.\nAnd a global skill's description must state its boundaries, because it loads into every session — the repo's\nown tests enforce that.\n\n## 5. Strip the provenance, always\n\nThe source repositories are private; this one is not. Everything in\n**`padosoft-skill-creator` §2** applies, and the gate runs before the commit:\n\n```bash\nmake privacy\n```\n\nNo dates, no customer or brand names, no people, no real tables, hosts or identifiers, no error messages\ncopied verbatim, no ticket or change numbers. **Keep the rule, drop the story** — and where a lesson only\nmakes sense with its story, the class of defect underneath it usually makes sense on its own.\n\n## 6. Record every decision\n\n```bash\npython3 scripts\u002Fharvest.py record \"monolith|docs\u002FLESSON.md\" covered --skill padosoft-query-performance\npython3 scripts\u002Fharvest.py record \"api|.claude\u002Frules\u002Frule-x.md\" rejected --note \"product-specific: names its own tables\"\npython3 scripts\u002Fharvest.py record \"mobile|.claude\u002Fskills\u002Fy\u002FSKILL.md\" pending --note \"good, needs a second source before promoting\"\n```\n\n`covered` requires a skill and `rejected` requires a reason — enforced, because a bare \"done\" is what makes\nthe next harvest re-read the same four hundred lines. `pending` is a legitimate state: a rule worth having\nthat is waiting for a second independent source.\n\n## 7. Close the loop\n\n`make catalog && make all`, update the changelog, commit, tag, release. Then state the numbers: sources\nscanned, findings, what was extended, what was created, what was rejected and why. A harvest that produced\nnothing is a valid outcome **and it is still recorded**, so the next one starts from the right place.\n\n## 8. After the harvest: the source and the skill are two layers, not two copies\n\nThis is the part that decides whether the whole loop is worth running. You harvested a rule **out of** a\nrepository, and now you install the package **into** that same repository. Without a further step you have\nthe rule twice: the agent loads both, they cost context, and the day the skill improves they disagree.\n\nThey are not duplicates. They are two layers:\n\n| Layer | Holds | Lives in |\n|---|---|---|\n| **The skill** | the class of defect — what goes wrong, why, what prevents it | the package, one copy for every project |\n| **The local rule** | the binding — what that class means *here*: the table, the helper, the path, the documented exception | the repository |\n\nSo the step after installing is to **thin the source, not delete it**. A four-hundred-line rule becomes\ntwenty lines that name the skill and keep only what is true about this repository.\n\n```bash\npython3 scripts\u002Fharvest.py adopt --id \u003Csource>     # exactly which files, and into which skill\n```\n\n```markdown\n\u003C!-- .claude\u002Frules\u002Frule-database-design.md, after adoption -->\n# Database design — local binding\n\nThe general rule is **`padosoft-database-design`** (installed with the `data` profile).\nThis file only records what is specific to this repository.\n\n- Index naming here is `idx_\u003Cabbrev>_\u003Ccol>_\u003Ccol>`.\n- The partition key helper is `App\\Support\\PartitionKey`.\n- Migration guards go through `App\\Support\\SchemaGuard`, not inline checks.\n- **Documented deviation:** the archive tables keep a redundant single-column index,\n  because the reporting tool cannot use a composite prefix. Revisit when it is replaced.\n```\n\nDelete it outright only when nothing project-specific is left. Keep the full local copy only in a repository\nthat cannot install the package at all — and write down that this is why.\n\n### Precedence, so a disagreement is never silent\n\n**The skill is the default. The local file may narrow it, add to it, or override it — and an override\ncarries a written reason.** A local rule that contradicts a skill without saying so is a bug in one of the\ntwo: either the project is a genuine exception (say so, in the binding) or the skill is wrong (fix the\nskill, for everybody).\n\n### Starting a new project\n\nInstall the profiles and **write no local rules at all**. The rules folder starts empty and grows only with\nbindings — a local file appears the first time this project needs to say something the general rule cannot\nknow. That is also the honest test of whether you needed it.\n\n### Adopting in a project that already has rules\n\nOnce, per repository, in this order:\n\n1. **Install** the profiles it needs.\n2. **Take the adoption list** above. For every file the ledger says is covered: thin it to its binding.\n3. **Leave what is not covered.** Those are candidates for the next harvest, not debt.\n4. **Resolve the contradictions explicitly.** A local rule that says the opposite of a skill is the most\n   valuable thing the adoption finds: one of the two is out of date, and until now nobody knew which.\n5. **Thin the derived instruction files too** — the copies for the other agents. Thinning one and not the\n   others is how you create the disagreement that\n   **`padosoft-agent-instructions-sync`** exists to prevent.\n\nDo it repository by repository, not all at once. Each one takes an hour and removes a source of drift\npermanently.\n\n## 9. The small loop: after every pull request or session\n\nThe harvest above is the periodic pass across repositories. The same idea at the scale of one piece of work\nis a habit, and it only works if it sits in the project's own always-loaded instructions (a line in the\nagent instruction file and in its rules), where nobody has to remember it.\n\nWhen a pull request or a session ends, before calling it done:\n\n1. **List what went wrong or had to be corrected**: a red check, a review finding, something done twice, a\n   trap in the environment.\n2. **Ask of each whether it is a class of defect that no rule covers yet.** If it is, add it to the rule that\n   would have prevented it, worded as a rule with its reason and without the story. If a rule already\n   covered it and was broken anyway, make that rule easier to find, or turn it into a check.\n3. **Say it in the pull request**: one line, \"Rules updated: …\", or \"No new rule: …\" and why.\n\nWhat a script can catch should become a script; text is the second choice (`padosoft-skill-audit`, G-FREE-02).\nA lessons file feeds this loop, and the periodic harvest then promotes what several projects learned alike.\n\n---\n\n## Gotchas\n\n- **The biggest file is usually the poorest.** A long lessons journal is mostly environment notes and review\n  chatter; the density is in the rule folders and the decision records.\n- **A rule with an identifier that nothing enforces means the control exists today and nothing protects it\n  tomorrow.** Worth harvesting, and worth saying so.\n- **Two sources with the same rule word-for-word are one source.** Shared ancestry is not convergence; check\n  whether one was copied from the other before counting it twice.\n- **The interesting material is often outside the rules folder** — a security directory, decision records,\n  a runbook. Adjust the globs per source once, rather than assuming the default found everything.\n- **A deleted source file does not mean the rule is wrong.** It means nothing in that repository defends it\n  any more, which is itself worth knowing.\n- **Re-running the scan after the work but before recording** shows the same findings. That is correct, and\n  it is the reason `record` exists as a separate step.\n\n## Checklist\n\n- [ ] Sources file current; every path resolves (problems section empty)\n- [ ] Work order read in full, including the *gone* group\n- [ ] Each finding judged: rule or note · would the agent get it wrong · class or instance\n- [ ] Extending preferred over creating; overlap with existing descriptions checked\n- [ ] Promotions justified by independent convergence, counted rather than felt\n- [ ] Provenance gate green before committing\n- [ ] Every finding recorded — covered with its skill, rejected with its reason, pending with what it waits for\n- [ ] Catalogue rebuilt, checks green, changelog written, release tagged\n- [ ] Adoption: every source file now covered by a skill thinned to its binding, derived instruction files included\n- [ ] After a pull request or session: its fixes and lessons read, and the rules that would have prevented them updated (§9)\n\n## Final report\n\n```\nHarvest \u003Cdate> — sources \u003Cn> · files read \u003Cn> · findings \u003Cn>\nExtended: \u003Cskill> ← \u003Csource|file> (\u003Cwhat the rule is>)\nCreated:  \u003Cskill> ← \u003Csources>\nPromoted: \u003Cskill> — converging sources: \u003Cn> (\u003Cwhich>)\nRejected: \u003Cn> — \u003Cthe categories, with an example each>\nPending:  \u003Cn> — waiting for \u003Cwhat>\nProvenance gate: clean\nRelease: \u003Ctag>\nAdoption pending: \u003Csource> — \u003Cn> files still holding a full copy\n```\n","# Know-how harvest\n\nEvery repository you run accumulates hard-won rules — in a lessons file, a rules folder, decision records,\nan internal skill. They stay there, benefiting one project. This is the loop that moves them into the shared\ncatalogue **without** re-reading everything every time, and without losing the record of what was decided.\n\n**The rule: harvest the rule, leave the story, and write down what you decided — including the rejections,\nbecause an undocumented rejection is re-mined every six months.**\n\n---\n\n## 0. Setup, once\n\n```bash\ncp .harvest-sources.example.json .harvest-sources.json   # then fill in your paths\n```\n\nBoth that file and `.harvest-ledger.json` are gitignored on purpose: **a list of the repositories you run is\nmetadata about your business**, and this repository is public. If more than one person harvests, keep them\nin a private companion repository or a secret rather than committing them here.\n\n## 1. The work order\n\n```bash\nmake harvest                 # or: python3 scripts\u002Fharvest.py scan\npython3 scripts\u002Fharvest.py scan --id monolith       # one source\npython3 scripts\u002Fharvest.py scan --json              # to drive another tool\n```\n\nIt prints four groups, and the current catalogue to match against:\n\n| Group | What it means |\n|---|---|\n| **New** | never considered — the bulk on a first run, a trickle afterwards |\n| **Changed** | reconsider; the work order says what you decided last time |\n| **Gone** | the source file is gone — check whether a skill still rests on it |\n| **Unreadable** | too large to read in one pass. Split it or exclude it **deliberately** |\n\nThe problems section is part of the output, not a warning: a source path that no longer resolves is a\nfinding. A harvest that silently skips a repository reports a clean catalogue — see\n**`padosoft-evidence-boundaries`**.\n\n**The scan never writes to the ledger.** A session that dies half way must not leave everything marked as\nprocessed.\n\n## 2. Read for the rule, not for the change\n\nFor each finding, open the file — for a changed one, the **added** material is what matters. Then ask, in\nthis order:\n\n1. **Is it a rule, or is it a note?** A rule states what goes wrong and what prevents it. A dated progress\n   entry, a local path, a vendor version, a review comment is a note: it was true for that repository on\n   that day and carries nothing.\n2. **Would the agent get it wrong without it?** If the model already does the right thing unprompted, the\n   rule adds context and not quality. That is the filter that keeps the catalogue small enough to be read.\n3. **Is it project-specific, or a class of defect?** *A named table must be registered in the resolver* is\n   specific; *a resolver that returns its input unchanged makes an incomplete registration\n   indistinguishable from a no-op* is the class. Harvest the class.\n\nMost findings fail one of the three. That is normal, and saying so explicitly is the point of the ledger.\n\n## 3. The four outcomes\n\n| Outcome | When | What you do |\n|---|---|---|\n| **Extend an existing skill** | the class already has a home | add the section; bump the version; say which skill in the ledger |\n| **Create a new skill** | a coherent domain nothing covers | **`padosoft-skill-creator`**, then register it in a package |\n| **Promote to global** | see §4 | move to `core`, state the boundaries in the description |\n| **Reject** | a note, already covered, too specific, or the agent gets it right anyway | record it **with the reason** |\n\nDefault to extending. A new skill that overlaps an existing one makes both trigger worse, and the\ndescription is what pays the price.\n\n## 4. Promotion is decided by independent convergence\n\nA rule belongs in `core` when **stacks that share no line of code arrived at it on their own**. Not when it\nfeels important, not when it appears in the biggest repository — when two or three unrelated sources\nproduce it independently.\n\n```bash\npython3 scripts\u002Fharvest.py status     # shows how many sources feed each skill\n```\n\nThat count is the evidence. There is no cap: if four sources converge on a fifth rule, it gets promoted.\nAnd a global skill's description must state its boundaries, because it loads into every session — the repo's\nown tests enforce that.\n\n## 5. Strip the provenance, always\n\nThe source repositories are private; this one is not. Everything in\n**`padosoft-skill-creator` §2** applies, and the gate runs before the commit:\n\n```bash\nmake privacy\n```\n\nNo dates, no customer or brand names, no people, no real tables, hosts or identifiers, no error messages\ncopied verbatim, no ticket or change numbers. **Keep the rule, drop the story** — and where a lesson only\nmakes sense with its story, the class of defect underneath it usually makes sense on its own.\n\n## 6. Record every decision\n\n```bash\npython3 scripts\u002Fharvest.py record \"monolith|docs\u002FLESSON.md\" covered --skill padosoft-query-performance\npython3 scripts\u002Fharvest.py record \"api|.claude\u002Frules\u002Frule-x.md\" rejected --note \"product-specific: names its own tables\"\npython3 scripts\u002Fharvest.py record \"mobile|.claude\u002Fskills\u002Fy\u002FSKILL.md\" pending --note \"good, needs a second source before promoting\"\n```\n\n`covered` requires a skill and `rejected` requires a reason — enforced, because a bare \"done\" is what makes\nthe next harvest re-read the same four hundred lines. `pending` is a legitimate state: a rule worth having\nthat is waiting for a second independent source.\n\n## 7. Close the loop\n\n`make catalog && make all`, update the changelog, commit, tag, release. Then state the numbers: sources\nscanned, findings, what was extended, what was created, what was rejected and why. A harvest that produced\nnothing is a valid outcome **and it is still recorded**, so the next one starts from the right place.\n\n## 8. After the harvest: the source and the skill are two layers, not two copies\n\nThis is the part that decides whether the whole loop is worth running. You harvested a rule **out of** a\nrepository, and now you install the package **into** that same repository. Without a further step you have\nthe rule twice: the agent loads both, they cost context, and the day the skill improves they disagree.\n\nThey are not duplicates. They are two layers:\n\n| Layer | Holds | Lives in |\n|---|---|---|\n| **The skill** | the class of defect — what goes wrong, why, what prevents it | the package, one copy for every project |\n| **The local rule** | the binding — what that class means *here*: the table, the helper, the path, the documented exception | the repository |\n\nSo the step after installing is to **thin the source, not delete it**. A four-hundred-line rule becomes\ntwenty lines that name the skill and keep only what is true about this repository.\n\n```bash\npython3 scripts\u002Fharvest.py adopt --id \u003Csource>     # exactly which files, and into which skill\n```\n\n```markdown\n\u003C!-- .claude\u002Frules\u002Frule-database-design.md, after adoption -->\n# Database design — local binding\n\nThe general rule is **`padosoft-database-design`** (installed with the `data` profile).\nThis file only records what is specific to this repository.\n\n- Index naming here is `idx_\u003Cabbrev>_\u003Ccol>_\u003Ccol>`.\n- The partition key helper is `App\\Support\\PartitionKey`.\n- Migration guards go through `App\\Support\\SchemaGuard`, not inline checks.\n- **Documented deviation:** the archive tables keep a redundant single-column index,\n  because the reporting tool cannot use a composite prefix. Revisit when it is replaced.\n```\n\nDelete it outright only when nothing project-specific is left. Keep the full local copy only in a repository\nthat cannot install the package at all — and write down that this is why.\n\n### Precedence, so a disagreement is never silent\n\n**The skill is the default. The local file may narrow it, add to it, or override it — and an override\ncarries a written reason.** A local rule that contradicts a skill without saying so is a bug in one of the\ntwo: either the project is a genuine exception (say so, in the binding) or the skill is wrong (fix the\nskill, for everybody).\n\n### Starting a new project\n\nInstall the profiles and **write no local rules at all**. The rules folder starts empty and grows only with\nbindings — a local file appears the first time this project needs to say something the general rule cannot\nknow. That is also the honest test of whether you needed it.\n\n### Adopting in a project that already has rules\n\nOnce, per repository, in this order:\n\n1. **Install** the profiles it needs.\n2. **Take the adoption list** above. For every file the ledger says is covered: thin it to its binding.\n3. **Leave what is not covered.** Those are candidates for the next harvest, not debt.\n4. **Resolve the contradictions explicitly.** A local rule that says the opposite of a skill is the most\n   valuable thing the adoption finds: one of the two is out of date, and until now nobody knew which.\n5. **Thin the derived instruction files too** — the copies for the other agents. Thinning one and not the\n   others is how you create the disagreement that\n   **`padosoft-agent-instructions-sync`** exists to prevent.\n\nDo it repository by repository, not all at once. Each one takes an hour and removes a source of drift\npermanently.\n\n## 9. The small loop: after every pull request or session\n\nThe harvest above is the periodic pass across repositories. The same idea at the scale of one piece of work\nis a habit, and it only works if it sits in the project's own always-loaded instructions (a line in the\nagent instruction file and in its rules), where nobody has to remember it.\n\nWhen a pull request or a session ends, before calling it done:\n\n1. **List what went wrong or had to be corrected**: a red check, a review finding, something done twice, a\n   trap in the environment.\n2. **Ask of each whether it is a class of defect that no rule covers yet.** If it is, add it to the rule that\n   would have prevented it, worded as a rule with its reason and without the story. If a rule already\n   covered it and was broken anyway, make that rule easier to find, or turn it into a check.\n3. **Say it in the pull request**: one line, \"Rules updated: …\", or \"No new rule: …\" and why.\n\nWhat a script can catch should become a script; text is the second choice (`padosoft-skill-audit`, G-FREE-02).\nA lessons file feeds this loop, and the periodic harvest then promotes what several projects learned alike.\n\n---\n\n## Gotchas\n\n- **The biggest file is usually the poorest.** A long lessons journal is mostly environment notes and review\n  chatter; the density is in the rule folders and the decision records.\n- **A rule with an identifier that nothing enforces means the control exists today and nothing protects it\n  tomorrow.** Worth harvesting, and worth saying so.\n- **Two sources with the same rule word-for-word are one source.** Shared ancestry is not convergence; check\n  whether one was copied from the other before counting it twice.\n- **The interesting material is often outside the rules folder** — a security directory, decision records,\n  a runbook. Adjust the globs per source once, rather than assuming the default found everything.\n- **A deleted source file does not mean the rule is wrong.** It means nothing in that repository defends it\n  any more, which is itself worth knowing.\n- **Re-running the scan after the work but before recording** shows the same findings. That is correct, and\n  it is the reason `record` exists as a separate step.\n\n## Checklist\n\n- [ ] Sources file current; every path resolves (problems section empty)\n- [ ] Work order read in full, including the *gone* group\n- [ ] Each finding judged: rule or note · would the agent get it wrong · class or instance\n- [ ] Extending preferred over creating; overlap with existing descriptions checked\n- [ ] Promotions justified by independent convergence, counted rather than felt\n- [ ] Provenance gate green before committing\n- [ ] Every finding recorded — covered with its skill, rejected with its reason, pending with what it waits for\n- [ ] Catalogue rebuilt, checks green, changelog written, release tagged\n- [ ] Adoption: every source file now covered by a skill thinned to its binding, derived instruction files included\n- [ ] After a pull request or session: its fixes and lessons read, and the rules that would have prevented them updated (§9)\n\n## Final report\n\n```\nHarvest \u003Cdate> — sources \u003Cn> · files read \u003Cn> · findings \u003Cn>\nExtended: \u003Cskill> ← \u003Csource|file> (\u003Cwhat the rule is>)\nCreated:  \u003Cskill> ← \u003Csources>\nPromoted: \u003Cskill> — converging sources: \u003Cn> (\u003Cwhich>)\nRejected: \u003Cn> — \u003Cthe categories, with an example each>\nPending:  \u003Cn> — waiting for \u003Cwhat>\nProvenance gate: clean\nRelease: \u003Ctag>\nAdoption pending: \u003Csource> — \u003Cn> files still holding a full copy\n```\n",{"name":5,"license":14,"metadata":23,"description":7,"compatibility":31},{"scope":24,"author":25,"summary":26,"version":27,"keywords":28,"profiles":29,"repository":30},"global","Padosoft","Harvest the rule, leave the story, and record what you decided so nobody mines it twice.","0.2.0","harvest, lessons, rules, catalogue, maintenance, promotion, ledger, know-how, periodic","core","https:\u002F\u002Fgithub.com\u002Fpadosoft\u002Fskills","Requires the repository's `harvest.py` (in its `scripts\u002F` folder) and read access to the source repositories. The sources and the ledger live in gitignored files, because a list of the repositories you run is metadata about your business.","main","8ff266762b08cab7f5be5551fa221309dc9db48b","skills\u002Fpadosoft-knowhow-harvest",0,"https:\u002F\u002Fgithub.com\u002Fpadosoft","https:\u002F\u002Fwww.padosoft.com","Software House","2026-10-19 01:02:17",{"source":8,"skillName":5,"commands":41,"prompt":51,"zipAvailable":52},[42,45,48],{"tool":43,"command":44},"npx","npx skills add padosoft\u002Fskills --skill padosoft-knowhow-harvest",{"tool":46,"command":47},"bunx","bunx skills add padosoft\u002Fskills --skill padosoft-knowhow-harvest",{"tool":49,"command":50},"pnpm","pnpm dlx skills add padosoft\u002Fskills --skill padosoft-knowhow-harvest","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fpadosoft\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fpadosoft-knowhow-harvest\u002FSKILL.md",true,[54],{"path":55,"resourceType":56,"resourceTypeDesc":57,"sizeBytes":58,"sourceHtmlUrl":59},"evals\u002Fqueries.json","REFERENCE","参考文档",1504,"https:\u002F\u002Fgithub.com\u002Fpadosoft\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fpadosoft-knowhow-harvest\u002Fevals\u002Fqueries.json",[61,68,74,81,88,94],{"skillNo":62,"name":63,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":67,"publishedAt":18},"01M7CZNKCTMWK7F2J6KKPAR8YM","report-opus","Writes the weekly status report from the team's tickets. Use when the user asks for the weekly report or a status summary. Do not use it for incident reports.","https:\u002F\u002Fgithub.com\u002Fpadosoft\u002Fskills\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskill-audit\u002Fproject\u002Fdot-claude\u002Fskills\u002Freport-opus\u002FSKILL.md",[],[],{"skillNo":69,"name":70,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":71,"categories":72,"keywords":73,"publishedAt":18},"01M7CZNFEN8XK6S8NC6KGREYGS","report-haiku","https:\u002F\u002Fgithub.com\u002Fpadosoft\u002Fskills\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskill-audit\u002Fproject\u002Fdot-claude\u002Fskills\u002Freport-haiku\u002FSKILL.md",[],[],{"skillNo":75,"name":76,"nameZh":6,"description":77,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":78,"categories":79,"keywords":80,"publishedAt":18},"01M7CZND182TKGGM7R4134ND00","kitt-stub","Builds the guided-tour reference for an admin page. Use when the user asks for the tour reference. Trigger \"tour\", \"reference\", \"guided tour\".","https:\u002F\u002Fgithub.com\u002Fpadosoft\u002Fskills\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskill-audit\u002Fproject\u002Fdot-claude\u002Fskills\u002Fkitt-stub\u002FSKILL.md",[],[],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":18},"01M7CZN9F2A6BFTFW5RSRNAGB1","good-skill","Generates the release notes of a project from its merged pull requests. Use this skill when the user asks for release notes, a changelog entry or a summary of what shipped since the last tag. Do not use it to write commit messages, that is the job of a commit skill.","https:\u002F\u002Fgithub.com\u002Fpadosoft\u002Fskills\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskill-audit\u002Fproject\u002Fdot-claude\u002Fskills\u002Fgood-skill\u002FSKILL.md",[],[],{"skillNo":89,"name":90,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"publishedAt":18},"01M7CZN6BJASPD4DH8DFR63GQF","broken-frontmatter","https:\u002F\u002Fgithub.com\u002Fpadosoft\u002Fskills\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskill-audit\u002Fproject\u002Fdot-claude\u002Fskills\u002Fbroken-frontmatter\u002FSKILL.md",[],[],{"skillNo":95,"name":96,"nameZh":6,"description":97,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":98,"categories":99,"keywords":100,"publishedAt":18},"01M7CZN498TRNC30GSYMMAZNXM","Bad_Skill","I can help you with reports.","https:\u002F\u002Fgithub.com\u002Fpadosoft\u002Fskills\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fskill-audit\u002Fproject\u002Fdot-claude\u002Fskills\u002Fbad-skill\u002FSKILL.md",[],[]]