[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7CJS3BXTZN6TFWSK0F6TNE6":3,"skill-same:01M7CJS3BXTZN6TFWSK0F6TNE6":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":26,"commitSha":27,"skillDirectory":28,"forksCount":29,"repositoryHtmlUrl":30,"repositoryHomepageUrl":6,"ownerHtmlUrl":31,"ownerBlogUrl":32,"ownerDescription":33,"updatedAt":34,"install":35,"resources":48},"01M7CJS3BXTZN6TFWSK0F6TNE6","chapter-writing",null,"Develop a specification chapter with the human through pair-programming, then teach what the work established.","waywardgeek\u002Fensemble","waywardgeek","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F350306?v=4","01M7CJR3NFF2YNXJEE7ST76W5P","Bill","01M7CJR3P8WHVZ9P2H4E735YMM",62,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fwaywardgeek\u002Fensemble\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fdev-driven-specs\u002Fchapter-writing\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: chapter-writing\ndescription: Develop a specification chapter with the human through pair-programming, then teach what the work established.\ntype: loadable\ndepends: coding visible-reasoning\n---\n\n# Write a chapter through development\n\nExplicitly load [coding](..\u002Fcoding\u002FSKILL.md) and\n[visible-reasoning](..\u002Fvisible-reasoning\u002FSKILL.md) unless already active. Do not\nassume `depends` activates them. If skill loading is unavailable, ask the human to\nadd their bodies to persistent instructions; reading them once is not equivalent.\n\nYou are the human's coding and writing collaborator. Develop the implementation\nand specification together in the same conversation, with the human steering.\nThese instructions contain the procedure; the dependencies supply architecture\nand explanation practices. Do not hand off the learning to hidden workers.\n\n## 1. Discuss the next useful increment\n\nDo not start with a book outline, chapter list, or interview to manufacture one.\nA plausible outline is not expertise. An expert can guide creation chapter by\nchapter without first transcribing their entire mental model. Someone developing\nexpertise cannot yet predict the final structure. Let the human guide the next\nincrement rather than prescribing the book ahead of the learning.\n\nAsk what to build next. Read relevant earlier chapters, inspect the starting code,\nand discuss what the result must demonstrate. If the human explicitly requests\nan outline, treat it as provisional, not a contract that later learning must obey.\n\nDraft only the **TL;DR and exercises**: enough behavior, boundaries and checks to\nbegin implementing. Development may correct this working specification. Do not\nwrite the body or leap ahead into unearned architecture.\n\n## 2. Pair-program the implementation and graders\n\nUse bounded experiments and small checked edits. Apply the loaded coding and\nvisible-reasoning skills: explain before acting, let the human steer, and report\nactual outcomes. Continue ordinary agreed work without per-command approval.\n\nBuild exercise graders alongside the implementation. Keep graders and private\nfixtures outside the solution tree, with a documented entry point accepting an\nindependently supplied implementation. Grade the agreed behavior, not hidden\nrequirements or incidental reference structure. Investigate failures and check\nthat important tests detect plausible defects rather than merely producing passes.\n\n## 3. Capture the learning while it happens\n\nKeep concise notes on decisions and reasons, ownership, alternatives, meaningful\nfailures, corrections, observed checks and open questions. State lessons in chat,\nnot only private notes. Update the TL;DR and exercises when the design changes,\npreserving why. Notes are raw material for teaching, not a new reporting system.\n\n## 4. Write the chapter from demonstrated experience\n\nOnce the implementation passes its exercise graders, read the notes and the\nproject's voice guide if it has one. Write the body from what actually worked:\n\n- Explain the problem and plain-language idea before the machinery.\n- Teach construction, Common vocabulary, relationships, lifecycle and reasons\n  coherently, using examples and meaningful failures rather than a patch diary.\n- Put exact interfaces, formats, errors and edge cases in labeled reference\n  material, with one authoritative home per requirement.\n- Keep the teaching, implementation and graders consistent. Repair original\n  teaching points rather than accumulating release-note appendices.\n\nWrite plainly and preserve useful reasons. Use actual experiences, quotations and\nresults; do not invent history or overstate evidence. Snippets should teach without\nhanding the student the complete target implementation.\n\n### Make the student's completion requirements explicit\n\nInclude **Exercises** or an equivalent clearly labeled assignment: what to implement,\nexactly which checks must pass, and how to recognize success. Give test\u002Fgrader\ncommands, working directory, prerequisites and how to supply the implementation.\nPrivate grader code must not conceal additional requirements.\n\nDistinguish required checks from optional experiments, including any agreed manual\nchecks. Unperformed checks are not passes. Verify the published commands against\nactual work; the student must not need our conversation to discover what “done” means.\n\n### Include “Take it for a Spin”\n\nGive the human a short, practical walkthrough of the chapter's actual behavior,\nseparate from the student's exercises. Use copy\u002Fpaste-ready commands or explicit\nUI steps, not a request to invent tests or merely run graders. State setup, starting\ndirectory, substitutions, expected sights or sounds, and stop\u002Fcleanup steps where\nneeded. Explain costs, credentials and external effects before those steps.\n\nUse safe, disposable examples where appropriate and check the instructions against\nthe implementation. The human should be able to kick the tires without reading\nsource or reconstructing our development conversation.\n\n## 5. Call the code reviewer exactly once\n\nAfter completing the initial implementation, passing its graders and writing the\nchapter body, call a separate agent with the\n[code-review skill](..\u002Fcodebook-code-reviewer\u002FSKILL.md). Give it the chapter,\nimplementation, tests and grader results. Ask for one consolidated review against\nthe stated requirements and coding rules, not new features or requirements.\n\nRead its findings, incorporate justified feedback and run affected checks yourself.\nDo not send the changes back for another review or recruit another reviewer. One\nreview, one incorporation pass, then move on. If a finding needs a consequential\nhuman decision, ask the human rather than starting a reviewer loop.\n\n## 6. Call the chapter reviewer exactly once\n\nNext call a separate agent with the\n[chapter-review skill](..\u002Fcodebook-text-reviewer\u002FSKILL.md). Give it the chapter and\nrelevant implementation evidence. Ask for one consolidated review of clarity,\naccuracy, teaching and consistency with the actual requirements.\n\nIncorporate its useful feedback yourself. Do not iterate with the reviewer, request\nre-review or replace it with another reviewer. Collect each reviewer's result and\nfinish its work before moving on. Review is assistance, not an endless search for\nsomething else to change.\n\n## 7. Launch the student and help it succeed\n\nLaunch a fresh student with the [spec-execution skill](..\u002Fspec-execution\u002FSKILL.md),\nthe completed chapter, accepted prerequisite code, an isolated solution workspace\nand the public grader commands. Tell it this is a chapter trial: implement the\nassignment, pass its graders and return feedback to the author. Do not add another\nstudent-owned review process. The student can read its own work and prerequisites,\nbut not our target implementation, earlier target solutions, private repair history,\ngrader source or private fixtures.\n\nBe helpful when the student asks questions. Clarify any ambiguity from the human's\nactual requirements; do not make it guess our intent. Keep notes on questions,\nanswers and teaching gaps for the chapter update. Do not invent requirements the\nhuman never stated or quietly supply the reference implementation.\n\nYou may modify the graders to help the student: correct bugs, ambiguous expectations,\nmisleading diagnostics or demands that do not belong to the agreed requirements.\nKeep those changes consistent with the chapter and tell the student what changed.\nThe student reruns the corrected graders. Do not preserve a bad grader merely because\nit already exists, or weaken a valid requirement just to manufacture a pass.\n\nWhen the student passes **100% of the required graders**, incorporate its feedback\nand your clarification notes into the chapter, then tell the human it passed.\nDescribe actual assistance, grader corrections and any remaining known limitations\nplainly. Useful quoted exchanges must match the saved wording, retain corrections\nand exclude private material. Clarification-assisted success is still useful success;\ndo not label it unassisted or claim the original page already contained every answer.\n\n**Never rerun a student that has passed all graders without discussing it with the\nhuman.** Do not invent new gates, launch another reviewer, or turn chapter polishing\ninto an automatic fresh-student campaign. If later edits raise a real concern, explain\nit to the human and let them decide what additional work is worthwhile.\n\n## 8. Human acceptance, then move on\n\nLet the human decide whether the student's work is good enough to advance. The\nhuman must also try the chapter's work using **Take it for a Spin** before accepting\nit; graders and agent demonstrations do not replace that test. If the walkthrough\nis blocked or unperformed, say so rather than claiming acceptance.\n\nAddress the human's feedback within the agreed scope. Preserve useful lessons and\ncontinue to the next chapter when the human accepts. Do not expand the task merely\nto collect more evidence.\n\n## Crossover closes the learning loop\n\nInclude a crossover chapter for the human's use of the system after book execution.\nThe generated implementation may pass its graders yet still need fixes to become\nusable. **Every crossover fix MUST be explained in prose in that chapter**: the\nobserved problem, cause or uncertainty, correction and its reason, and actual checks.\nRecord each lesson as the work happens; code changes and private notes are not enough.\n\nBefore executing the book again, use that record to repair the chapters that should\nhave taught how to avoid each issue, including exercises and graders when needed.\nRecord where the lessons went. Crossover prose preserves experience; the updated\nconstruction chapters turn it into better instructions for the next execution.\n","# Write a chapter through development\n\nExplicitly load [coding](..\u002Fcoding\u002FSKILL.md) and\n[visible-reasoning](..\u002Fvisible-reasoning\u002FSKILL.md) unless already active. Do not\nassume `depends` activates them. If skill loading is unavailable, ask the human to\nadd their bodies to persistent instructions; reading them once is not equivalent.\n\nYou are the human's coding and writing collaborator. Develop the implementation\nand specification together in the same conversation, with the human steering.\nThese instructions contain the procedure; the dependencies supply architecture\nand explanation practices. Do not hand off the learning to hidden workers.\n\n## 1. Discuss the next useful increment\n\nDo not start with a book outline, chapter list, or interview to manufacture one.\nA plausible outline is not expertise. An expert can guide creation chapter by\nchapter without first transcribing their entire mental model. Someone developing\nexpertise cannot yet predict the final structure. Let the human guide the next\nincrement rather than prescribing the book ahead of the learning.\n\nAsk what to build next. Read relevant earlier chapters, inspect the starting code,\nand discuss what the result must demonstrate. If the human explicitly requests\nan outline, treat it as provisional, not a contract that later learning must obey.\n\nDraft only the **TL;DR and exercises**: enough behavior, boundaries and checks to\nbegin implementing. Development may correct this working specification. Do not\nwrite the body or leap ahead into unearned architecture.\n\n## 2. Pair-program the implementation and graders\n\nUse bounded experiments and small checked edits. Apply the loaded coding and\nvisible-reasoning skills: explain before acting, let the human steer, and report\nactual outcomes. Continue ordinary agreed work without per-command approval.\n\nBuild exercise graders alongside the implementation. Keep graders and private\nfixtures outside the solution tree, with a documented entry point accepting an\nindependently supplied implementation. Grade the agreed behavior, not hidden\nrequirements or incidental reference structure. Investigate failures and check\nthat important tests detect plausible defects rather than merely producing passes.\n\n## 3. Capture the learning while it happens\n\nKeep concise notes on decisions and reasons, ownership, alternatives, meaningful\nfailures, corrections, observed checks and open questions. State lessons in chat,\nnot only private notes. Update the TL;DR and exercises when the design changes,\npreserving why. Notes are raw material for teaching, not a new reporting system.\n\n## 4. Write the chapter from demonstrated experience\n\nOnce the implementation passes its exercise graders, read the notes and the\nproject's voice guide if it has one. Write the body from what actually worked:\n\n- Explain the problem and plain-language idea before the machinery.\n- Teach construction, Common vocabulary, relationships, lifecycle and reasons\n  coherently, using examples and meaningful failures rather than a patch diary.\n- Put exact interfaces, formats, errors and edge cases in labeled reference\n  material, with one authoritative home per requirement.\n- Keep the teaching, implementation and graders consistent. Repair original\n  teaching points rather than accumulating release-note appendices.\n\nWrite plainly and preserve useful reasons. Use actual experiences, quotations and\nresults; do not invent history or overstate evidence. Snippets should teach without\nhanding the student the complete target implementation.\n\n### Make the student's completion requirements explicit\n\nInclude **Exercises** or an equivalent clearly labeled assignment: what to implement,\nexactly which checks must pass, and how to recognize success. Give test\u002Fgrader\ncommands, working directory, prerequisites and how to supply the implementation.\nPrivate grader code must not conceal additional requirements.\n\nDistinguish required checks from optional experiments, including any agreed manual\nchecks. Unperformed checks are not passes. Verify the published commands against\nactual work; the student must not need our conversation to discover what “done” means.\n\n### Include “Take it for a Spin”\n\nGive the human a short, practical walkthrough of the chapter's actual behavior,\nseparate from the student's exercises. Use copy\u002Fpaste-ready commands or explicit\nUI steps, not a request to invent tests or merely run graders. State setup, starting\ndirectory, substitutions, expected sights or sounds, and stop\u002Fcleanup steps where\nneeded. Explain costs, credentials and external effects before those steps.\n\nUse safe, disposable examples where appropriate and check the instructions against\nthe implementation. The human should be able to kick the tires without reading\nsource or reconstructing our development conversation.\n\n## 5. Call the code reviewer exactly once\n\nAfter completing the initial implementation, passing its graders and writing the\nchapter body, call a separate agent with the\n[code-review skill](..\u002Fcodebook-code-reviewer\u002FSKILL.md). Give it the chapter,\nimplementation, tests and grader results. Ask for one consolidated review against\nthe stated requirements and coding rules, not new features or requirements.\n\nRead its findings, incorporate justified feedback and run affected checks yourself.\nDo not send the changes back for another review or recruit another reviewer. One\nreview, one incorporation pass, then move on. If a finding needs a consequential\nhuman decision, ask the human rather than starting a reviewer loop.\n\n## 6. Call the chapter reviewer exactly once\n\nNext call a separate agent with the\n[chapter-review skill](..\u002Fcodebook-text-reviewer\u002FSKILL.md). Give it the chapter and\nrelevant implementation evidence. Ask for one consolidated review of clarity,\naccuracy, teaching and consistency with the actual requirements.\n\nIncorporate its useful feedback yourself. Do not iterate with the reviewer, request\nre-review or replace it with another reviewer. Collect each reviewer's result and\nfinish its work before moving on. Review is assistance, not an endless search for\nsomething else to change.\n\n## 7. Launch the student and help it succeed\n\nLaunch a fresh student with the [spec-execution skill](..\u002Fspec-execution\u002FSKILL.md),\nthe completed chapter, accepted prerequisite code, an isolated solution workspace\nand the public grader commands. Tell it this is a chapter trial: implement the\nassignment, pass its graders and return feedback to the author. Do not add another\nstudent-owned review process. The student can read its own work and prerequisites,\nbut not our target implementation, earlier target solutions, private repair history,\ngrader source or private fixtures.\n\nBe helpful when the student asks questions. Clarify any ambiguity from the human's\nactual requirements; do not make it guess our intent. Keep notes on questions,\nanswers and teaching gaps for the chapter update. Do not invent requirements the\nhuman never stated or quietly supply the reference implementation.\n\nYou may modify the graders to help the student: correct bugs, ambiguous expectations,\nmisleading diagnostics or demands that do not belong to the agreed requirements.\nKeep those changes consistent with the chapter and tell the student what changed.\nThe student reruns the corrected graders. Do not preserve a bad grader merely because\nit already exists, or weaken a valid requirement just to manufacture a pass.\n\nWhen the student passes **100% of the required graders**, incorporate its feedback\nand your clarification notes into the chapter, then tell the human it passed.\nDescribe actual assistance, grader corrections and any remaining known limitations\nplainly. Useful quoted exchanges must match the saved wording, retain corrections\nand exclude private material. Clarification-assisted success is still useful success;\ndo not label it unassisted or claim the original page already contained every answer.\n\n**Never rerun a student that has passed all graders without discussing it with the\nhuman.** Do not invent new gates, launch another reviewer, or turn chapter polishing\ninto an automatic fresh-student campaign. If later edits raise a real concern, explain\nit to the human and let them decide what additional work is worthwhile.\n\n## 8. Human acceptance, then move on\n\nLet the human decide whether the student's work is good enough to advance. The\nhuman must also try the chapter's work using **Take it for a Spin** before accepting\nit; graders and agent demonstrations do not replace that test. If the walkthrough\nis blocked or unperformed, say so rather than claiming acceptance.\n\nAddress the human's feedback within the agreed scope. Preserve useful lessons and\ncontinue to the next chapter when the human accepts. Do not expand the task merely\nto collect more evidence.\n\n## Crossover closes the learning loop\n\nInclude a crossover chapter for the human's use of the system after book execution.\nThe generated implementation may pass its graders yet still need fixes to become\nusable. **Every crossover fix MUST be explained in prose in that chapter**: the\nobserved problem, cause or uncertainty, correction and its reason, and actual checks.\nRecord each lesson as the work happens; code changes and private notes are not enough.\n\nBefore executing the book again, use that record to repair the chapters that should\nhave taught how to avoid each issue, including exercises and graders when needed.\nRecord where the lessons went. Crossover prose preserves experience; the updated\nconstruction chapters turn it into better instructions for the next execution.\n",{"name":5,"type":24,"depends":25,"description":7},"loadable","coding visible-reasoning","main","bb5a58aa1bf8c450d7f75b51f9a7e7d064914e10","agent\u002Fskills\u002Fdev-driven-specs\u002Fchapter-writing",4,"https:\u002F\u002Fgithub.com\u002Fwaywardgeek\u002Fensemble","https:\u002F\u002Fgithub.com\u002Fwaywardgeek","coderhapsody.ai","Wrote about 2 million lines of code\r\nWrote CodeRhapsody.ai\r\nFounded ViASIC\r\nListen at 600 wpm\r\nAuthor on 26 patents\r\nHuge geek for waaaay to long","2026-11-15 02:00:29",{"source":8,"skillName":5,"commands":36,"prompt":46,"zipAvailable":47},[37,40,43],{"tool":38,"command":39},"npx","npx skills add waywardgeek\u002Fensemble --skill chapter-writing",{"tool":41,"command":42},"bunx","bunx skills add waywardgeek\u002Fensemble --skill chapter-writing",{"tool":44,"command":45},"pnpm","pnpm dlx skills add waywardgeek\u002Fensemble --skill chapter-writing","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fwaywardgeek\u002Fensemble\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fdev-driven-specs\u002Fchapter-writing\u002FSKILL.md",true,[],[50,57,64,71,78,85],{"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},"01M7CKE3JEG4SFEK3PXS1BYSGJ","web-search","Search the web and fetch pages through Firecrawl","https:\u002F\u002Fgithub.com\u002Fwaywardgeek\u002Fensemble\u002Fblob\u002Fmain\u002Fsolutions\u002Fch27\u002FSkills\u002Fweb-search\u002FSKILL.md",[],[],{"skillNo":58,"name":59,"nameZh":6,"description":60,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":61,"categories":62,"keywords":63,"publishedAt":19},"01M7CKE1HFV0SBR89CACS4G8Z3","gui-debug","Inspect and test the connected Ensemble page with explicit browser tools.","https:\u002F\u002Fgithub.com\u002Fwaywardgeek\u002Fensemble\u002Fblob\u002Fmain\u002Fsolutions\u002Fch27\u002FSkills\u002Fgui-debug\u002FSKILL.md",[],[],{"skillNo":65,"name":66,"nameZh":6,"description":67,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":68,"categories":69,"keywords":70,"publishedAt":19},"01M7CKE00AQ9RF89EP7B7A40RT","goals","Keep an editable goal in request instructions and optionally resume after an idle delay.","https:\u002F\u002Fgithub.com\u002Fwaywardgeek\u002Fensemble\u002Fblob\u002Fmain\u002Fsolutions\u002Fch27\u002FSkills\u002Fgoals\u002FSKILL.md",[],[],{"skillNo":72,"name":73,"nameZh":6,"description":74,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":75,"categories":76,"keywords":77,"publishedAt":19},"01M7CKDWCTT95VJ8YN4NG3VYXD","full","Explicit full-capability primary for trusted local exercises.","https:\u002F\u002Fgithub.com\u002Fwaywardgeek\u002Fensemble\u002Fblob\u002Fmain\u002Fsolutions\u002Fch27\u002FSkills\u002Ffull\u002FSKILL.md",[],[],{"skillNo":79,"name":80,"nameZh":6,"description":81,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":82,"categories":83,"keywords":84,"publishedAt":19},"01M7CKDRBEQXQSFGTFPZPQGJTV","ensemble","Read, inspect, and explain before changing a workspace.","https:\u002F\u002Fgithub.com\u002Fwaywardgeek\u002Fensemble\u002Fblob\u002Fmain\u002Fsolutions\u002Fch27\u002FSkills\u002Fensemble\u002FSKILL.md",[],[],{"skillNo":86,"name":87,"nameZh":6,"description":88,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":89,"categories":90,"keywords":91,"publishedAt":19},"01M7CKDPCVR1JX39FGHQZHZMKB","code-tools","Edit files and run independent background jobs.","https:\u002F\u002Fgithub.com\u002Fwaywardgeek\u002Fensemble\u002Fblob\u002Fmain\u002Fsolutions\u002Fch27\u002FSkills\u002Fcode-tools\u002FSKILL.md",[],[]]