[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7CZC2AXMMC2A4993P17VR90":3,"skill-same:01M7CZC2AXMMC2A4993P17VR90":56},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":14,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":29,"install":30,"resources":43},"01M7CZC2AXMMC2A4993P17VR90","sweep-tests",null,"Prune or audit a test suite with mutation proof, as its own announced task on a green suite. Invoke it by name when tests break on refactors that change no behavior, coverage is high but defects still ship, or production code carries seams only tests use.","fern-works\u002Fgreenline","fern-works","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F330252531?v=4","01M7CZ6VNSYC4MB0XZS6BWJRX3","fernworks","01M7CZ6VPZYZWBPRG404SRBNA9",0,"MIT","https:\u002F\u002Fgithub.com\u002Ffern-works\u002Fgreenline\u002Fblob\u002Fmain\u002Fcorpus\u002Fskills\u002Fsweep-tests\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: \"sweep-tests\"\ndescription: \"Prune or audit a test suite with mutation proof, as its own announced task on a green suite. Invoke it by name when tests break on refactors that change no behavior, coverage is high but defects still ship, or production code carries seams only tests use.\"\n---\n\n# Test value sweep\n\nSweep tests is the start for a test suite that needs pruning or an audit. It fires by name, when tests break on refactors that change no behavior, when coverage is high but defects still ship, or when production code carries seams only tests use, and it runs on a green suite as a change of its own, never inside another change's build: propose its mode and wait, as AGENTS.md says, open its record with a `## start` section naming sweep-tests, and run the line from `frame`, the sweep itself being the change's build. Propose regular or above, since the sweep's result is reviewed by an agent other than the sweeper only where the review step runs independent; where the owner chooses quick, no review runs, and the close section names that limit. A red check goes to diagnose. It deletes nothing it has not proved by mutation, adds no test, and never certifies its own review without naming that limit.\n\nGreenline's paths and commands for this skill stand on its references page, [references\u002Fgreenline.md](references\u002Fgreenline.md): read it when the work lands in the workspace.\n\nA test must justify its presence. A test that changes when the implementation\nchanges, while the behavior stays the same, asserts implementation. Delete it.\n\nThen simplify the source. Seams in production code often exist only because\nthose tests demanded them. When the test goes, the seam goes with it.\n\n## The gate\n\n**Run this only when the suite is green.**\n\nA red check makes test deletion fraud, not maintenance. When a check is red,\nrepair the source through diagnose, and restart the sweep from the top\nonce the suite is green again. You may not touch a test at all until the suite\nis green again.\n\nThis skill is a separate change with its own change record.\nAnnounce it. Never run it inside another change's build.\n\n## Prove each deletion\n\nYou may not read a test and then declare it worthless. That judgment certifies\nitself, and it always says yes. Run the mutation check instead.\n\n**If the repository already has a mutation runner, use it.** Look for Stryker,\nmutmut, go-mutesting, PIT, or a `test:mutation` script. Its configuration holds\nknowledge this skill does not have, such as which files repay the run and which\ngenerate thousands of worthless mutants.\n\n**The mutation scope must cover the subject file of every test you delete in the\nbatch.** A test whose subject sits outside the scope is never proved, however\ngreen the score looks. Widen the scope, or split the batch.\n\nOnly when the repository has none, use the tool that ships here, `bin\u002Fmutate.mjs`\nin this skill's installed directory. Find it with the glob\n`**\u002Fsweep-tests\u002Fbin\u002Fmutate.mjs` under the installed skills rather than assuming\na harness:\n\n```bash\nnode \u003Cskill-dir>\u002Fbin\u002Fmutate.mjs --file src\u002Fthing.js --fn theFunction\n```\n\nEither tool breaks the source on purpose, one change at a time, and runs the\nsuite after each change. A mutant that survives is a behavior that no test\ncovers.\n\nValidate the measurement before judging a survivor. The bundled tool masks\nstrings and comments for its comparison and numeric operators, but its raw-text\nternary operator can still swap a quoted ternary example inside a comment, and\nits \"defect your suite ships\" line is a candidate, not a diagnosis. A mutation\nthat changes no executable behavior, or whose validity is unresolved, is an\ninvalid measurement, not a shipped defect: the run proves no deletion, so keep\nthe tests and seams, record the invalid measurement with the sweep's evidence,\nand use an already available capable runner only within the authorized scope.\nNever add a test for a comment, and never drop a baseline mutant to make the\nscore pass.\n\nA survivor inside error-handling code is the sweep's highest-value finding.\nCode that has not run does not work, and error handlers run rarest of all: a\nstudy of production failures in distributed data-intensive systems (Yuan et\nal., OSDI 2014) found 92% of catastrophic failures came from incorrect handling\nof non-fatal errors the software explicitly signaled. A test that exercises an\nerror path is never pruned as redundant on coverage grounds alone; weigh its\ndeletion against what its mutants proved.\n\n**Take the report before you delete anything. Take it again at the end.**\n\nCompare mutants by identity, not by number. Three failures hide behind an\nunchanged total:\n\n1. **A trade.** Two mutants swap status and the total holds.\n2. **A disappearance.** You removed the source a mutant lived in, so the mutant\n   is gone. Nothing moved from killed to survived, and the behaviour is now\n   untested. This is the one that catches seam removal.\n3. **A scope change.** The second run covered fewer files than the first.\n\nSo: record every baseline mutant identifier before you start, with the source\nhash and the exact test command. At the end, **every baseline mutant must still\nexist and must still be killed.** A missing mutant fails the sweep exactly as a\nsurvived one does.\n\nThe suite stays green throughout, so green proves nothing here.\n\nThe score is not proof either. It shows that the mutants this runner generated,\nunder this operator set, over this file set, kept their status. Behaviour no\noperator reaches is invisible to it, and a dynamic caller found only through a\nconfiguration string or a plugin registry is invisible to your grep. **A sweep\nwithout a score has proved nothing. A sweep with one has proved something\nnarrow.** Say which.\n\nWrite the before-report and the after-report to different paths in the\nsweep change's folder. A runner\nthat overwrites its own baseline leaves you unable to find the mutant you lost.\n\n**A runner that reports per-test identities names one killer for each mutant:\nthe first test that reached it.** That is run order, not ownership. The bundled\n`mutate.mjs` reports suite-level killed and survived outcomes and names no\nkiller; do not invent that attribution, and its silence is not evidence that a\ntest is redundant. A test credited with no kills can still be the only real net\nunder a mutant that another test happens to reach first. Use the report to\nshortlist candidates. Never use it as proof. The rerun is the proof.\n\n### Delete in batches\n\nTime one scoped run before you plan the sweep. The cost ranges from seconds to\nmany minutes, so whether one run for each candidate is affordable depends on the\nrepository. When it is not, delete a batch and run once.\n\n- The score holds. The whole batch stands.\n- The score falls. One test in the batch was the only killer of a mutant.\n  Restore the batch, halve it, and run again. Repeat until you find the test\n  that matters, then keep it and delete the rest.\n\nA batch proof is weaker than a proof for each test. It shows that no deletion\nwas the unique killer of a mutant. That is enough, and it is affordable.\n\n## Delete\n\n- Tests that assert that a call happened, when the call is not the contract\n- Tests whose assertions restate the source line above them\n- Snapshot tests that you regenerate instead of read\n- Tests that exercise only a mock\n- Repeated coverage of one behavior across many cases\n- Tests for a private function that a public function already reaches\n\n## Keep\n\n- The only test that covers a behavior, however ugly it is\n- Call assertions on a unit whose whole job is to drive other units. There the\n  call and its arguments are the result, and the mutation score will show it\n- Regression tests that name a defect or an issue\n- Contract tests at a boundary that you do not own\n- Property tests and fuzz tests\n\n## Remove the seams\n\nA deleted test may have been the only caller of a seam. Grep for each one.\n\n- An interface with one implementation\n- A constructor parameter that only a test supplies\n- An exported symbol that only tests import\n- A mock or fake, and the hook it plugs into\n- A flag that only a test reads\n\nDelete the seam, or inline it. Follow each simplification to its end. When you\nremove a seam and a wrapper becomes a pass-through, remove the wrapper too.\n\n**A seam can have one caller, and that caller can be a test in another file.\nLeave that seam alone.** Report it as a candidate for a later sweep. A sweep that\nreaches past its own scope breaks files that nobody asked you to touch.\n\n**A grep does not find every caller.** Dependency injection tokens, plugin\nregistries, configuration strings and generated registrations all reference a\nsymbol without naming it in a way `rg` matches. Removing such a seam keeps every\nunit test green, removes the mutants that lived in it, and fails in production.\nWhen a symbol is exported across a package boundary, treat the grep as\ninconclusive and leave it.\n\n## Report\n\nClose with the two scores and one table.\n\n```\nTests: 26 -> 9 (-65%)   Mutation score: 16\u002F18 -> 16\u002F18 (no change)\n```\n\nThe table lists each test you deleted and each seam you removed. Prose is not\nthe report.\n\nThe baselines, the identified survivors and kills, both reports, the table and\nthe resulting checks stay with the sweep's change in its folder, and\nacceptance is judged against those measurements: an aggregate score alone\ncannot prove that each baseline mutant remains killed. The sweep is\nimplementation work. Commit its result with the change's trailer for its\nreview step, like any build; no review runs only\nwhere the owner chose quick, and the close section names that limit.\n\n## Don't\n\n- Delete a test to turn a red check green\n- Delete a test that you did not prove by mutation\n- Rewrite a test and count that as a sweep\n- Add a test in this pass\n- Change anything outside the test suite and its seams\n\n## It's working if\n\n- Every deletion has a mutation report before it and one after it.\n- Both reports name the same mutants, not just the same total.\n- The close is two scores and a table. Prose is not the report.\n- Seams disappear in the same pass as the tests that demanded them.\n- A sweep that cannot prove a deletion says so, and keeps the test.\n- You deleted nothing while a check showed red.\n\n## Handoff\n\nConsumes: a green suite and the request to sweep it\nProduces: the before and after mutation results at different paths, the deletion and seam table, and the baselines and kills, in the change's folder under .greenline\u002Fledger\u002F\u003Cid>\u002F; the sweep committed as the change's build\nNext: frame, under the mode the owner approves; the sweep is the change's build, and review and verify follow at the depths its mode gives\n","# Test value sweep\n\nSweep tests is the start for a test suite that needs pruning or an audit. It fires by name, when tests break on refactors that change no behavior, when coverage is high but defects still ship, or when production code carries seams only tests use, and it runs on a green suite as a change of its own, never inside another change's build: propose its mode and wait, as AGENTS.md says, open its record with a `## start` section naming sweep-tests, and run the line from `frame`, the sweep itself being the change's build. Propose regular or above, since the sweep's result is reviewed by an agent other than the sweeper only where the review step runs independent; where the owner chooses quick, no review runs, and the close section names that limit. A red check goes to diagnose. It deletes nothing it has not proved by mutation, adds no test, and never certifies its own review without naming that limit.\n\nGreenline's paths and commands for this skill stand on its references page, [references\u002Fgreenline.md](references\u002Fgreenline.md): read it when the work lands in the workspace.\n\nA test must justify its presence. A test that changes when the implementation\nchanges, while the behavior stays the same, asserts implementation. Delete it.\n\nThen simplify the source. Seams in production code often exist only because\nthose tests demanded them. When the test goes, the seam goes with it.\n\n## The gate\n\n**Run this only when the suite is green.**\n\nA red check makes test deletion fraud, not maintenance. When a check is red,\nrepair the source through diagnose, and restart the sweep from the top\nonce the suite is green again. You may not touch a test at all until the suite\nis green again.\n\nThis skill is a separate change with its own change record.\nAnnounce it. Never run it inside another change's build.\n\n## Prove each deletion\n\nYou may not read a test and then declare it worthless. That judgment certifies\nitself, and it always says yes. Run the mutation check instead.\n\n**If the repository already has a mutation runner, use it.** Look for Stryker,\nmutmut, go-mutesting, PIT, or a `test:mutation` script. Its configuration holds\nknowledge this skill does not have, such as which files repay the run and which\ngenerate thousands of worthless mutants.\n\n**The mutation scope must cover the subject file of every test you delete in the\nbatch.** A test whose subject sits outside the scope is never proved, however\ngreen the score looks. Widen the scope, or split the batch.\n\nOnly when the repository has none, use the tool that ships here, `bin\u002Fmutate.mjs`\nin this skill's installed directory. Find it with the glob\n`**\u002Fsweep-tests\u002Fbin\u002Fmutate.mjs` under the installed skills rather than assuming\na harness:\n\n```bash\nnode \u003Cskill-dir>\u002Fbin\u002Fmutate.mjs --file src\u002Fthing.js --fn theFunction\n```\n\nEither tool breaks the source on purpose, one change at a time, and runs the\nsuite after each change. A mutant that survives is a behavior that no test\ncovers.\n\nValidate the measurement before judging a survivor. The bundled tool masks\nstrings and comments for its comparison and numeric operators, but its raw-text\nternary operator can still swap a quoted ternary example inside a comment, and\nits \"defect your suite ships\" line is a candidate, not a diagnosis. A mutation\nthat changes no executable behavior, or whose validity is unresolved, is an\ninvalid measurement, not a shipped defect: the run proves no deletion, so keep\nthe tests and seams, record the invalid measurement with the sweep's evidence,\nand use an already available capable runner only within the authorized scope.\nNever add a test for a comment, and never drop a baseline mutant to make the\nscore pass.\n\nA survivor inside error-handling code is the sweep's highest-value finding.\nCode that has not run does not work, and error handlers run rarest of all: a\nstudy of production failures in distributed data-intensive systems (Yuan et\nal., OSDI 2014) found 92% of catastrophic failures came from incorrect handling\nof non-fatal errors the software explicitly signaled. A test that exercises an\nerror path is never pruned as redundant on coverage grounds alone; weigh its\ndeletion against what its mutants proved.\n\n**Take the report before you delete anything. Take it again at the end.**\n\nCompare mutants by identity, not by number. Three failures hide behind an\nunchanged total:\n\n1. **A trade.** Two mutants swap status and the total holds.\n2. **A disappearance.** You removed the source a mutant lived in, so the mutant\n   is gone. Nothing moved from killed to survived, and the behaviour is now\n   untested. This is the one that catches seam removal.\n3. **A scope change.** The second run covered fewer files than the first.\n\nSo: record every baseline mutant identifier before you start, with the source\nhash and the exact test command. At the end, **every baseline mutant must still\nexist and must still be killed.** A missing mutant fails the sweep exactly as a\nsurvived one does.\n\nThe suite stays green throughout, so green proves nothing here.\n\nThe score is not proof either. It shows that the mutants this runner generated,\nunder this operator set, over this file set, kept their status. Behaviour no\noperator reaches is invisible to it, and a dynamic caller found only through a\nconfiguration string or a plugin registry is invisible to your grep. **A sweep\nwithout a score has proved nothing. A sweep with one has proved something\nnarrow.** Say which.\n\nWrite the before-report and the after-report to different paths in the\nsweep change's folder. A runner\nthat overwrites its own baseline leaves you unable to find the mutant you lost.\n\n**A runner that reports per-test identities names one killer for each mutant:\nthe first test that reached it.** That is run order, not ownership. The bundled\n`mutate.mjs` reports suite-level killed and survived outcomes and names no\nkiller; do not invent that attribution, and its silence is not evidence that a\ntest is redundant. A test credited with no kills can still be the only real net\nunder a mutant that another test happens to reach first. Use the report to\nshortlist candidates. Never use it as proof. The rerun is the proof.\n\n### Delete in batches\n\nTime one scoped run before you plan the sweep. The cost ranges from seconds to\nmany minutes, so whether one run for each candidate is affordable depends on the\nrepository. When it is not, delete a batch and run once.\n\n- The score holds. The whole batch stands.\n- The score falls. One test in the batch was the only killer of a mutant.\n  Restore the batch, halve it, and run again. Repeat until you find the test\n  that matters, then keep it and delete the rest.\n\nA batch proof is weaker than a proof for each test. It shows that no deletion\nwas the unique killer of a mutant. That is enough, and it is affordable.\n\n## Delete\n\n- Tests that assert that a call happened, when the call is not the contract\n- Tests whose assertions restate the source line above them\n- Snapshot tests that you regenerate instead of read\n- Tests that exercise only a mock\n- Repeated coverage of one behavior across many cases\n- Tests for a private function that a public function already reaches\n\n## Keep\n\n- The only test that covers a behavior, however ugly it is\n- Call assertions on a unit whose whole job is to drive other units. There the\n  call and its arguments are the result, and the mutation score will show it\n- Regression tests that name a defect or an issue\n- Contract tests at a boundary that you do not own\n- Property tests and fuzz tests\n\n## Remove the seams\n\nA deleted test may have been the only caller of a seam. Grep for each one.\n\n- An interface with one implementation\n- A constructor parameter that only a test supplies\n- An exported symbol that only tests import\n- A mock or fake, and the hook it plugs into\n- A flag that only a test reads\n\nDelete the seam, or inline it. Follow each simplification to its end. When you\nremove a seam and a wrapper becomes a pass-through, remove the wrapper too.\n\n**A seam can have one caller, and that caller can be a test in another file.\nLeave that seam alone.** Report it as a candidate for a later sweep. A sweep that\nreaches past its own scope breaks files that nobody asked you to touch.\n\n**A grep does not find every caller.** Dependency injection tokens, plugin\nregistries, configuration strings and generated registrations all reference a\nsymbol without naming it in a way `rg` matches. Removing such a seam keeps every\nunit test green, removes the mutants that lived in it, and fails in production.\nWhen a symbol is exported across a package boundary, treat the grep as\ninconclusive and leave it.\n\n## Report\n\nClose with the two scores and one table.\n\n```\nTests: 26 -> 9 (-65%)   Mutation score: 16\u002F18 -> 16\u002F18 (no change)\n```\n\nThe table lists each test you deleted and each seam you removed. Prose is not\nthe report.\n\nThe baselines, the identified survivors and kills, both reports, the table and\nthe resulting checks stay with the sweep's change in its folder, and\nacceptance is judged against those measurements: an aggregate score alone\ncannot prove that each baseline mutant remains killed. The sweep is\nimplementation work. Commit its result with the change's trailer for its\nreview step, like any build; no review runs only\nwhere the owner chose quick, and the close section names that limit.\n\n## Don't\n\n- Delete a test to turn a red check green\n- Delete a test that you did not prove by mutation\n- Rewrite a test and count that as a sweep\n- Add a test in this pass\n- Change anything outside the test suite and its seams\n\n## It's working if\n\n- Every deletion has a mutation report before it and one after it.\n- Both reports name the same mutants, not just the same total.\n- The close is two scores and a table. Prose is not the report.\n- Seams disappear in the same pass as the tests that demanded them.\n- A sweep that cannot prove a deletion says so, and keeps the test.\n- You deleted nothing while a check showed red.\n\n## Handoff\n\nConsumes: a green suite and the request to sweep it\nProduces: the before and after mutation results at different paths, the deletion and seam table, and the baselines and kills, in the change's folder under .greenline\u002Fledger\u002F\u003Cid>\u002F; the sweep committed as the change's build\nNext: frame, under the mode the owner approves; the sweep is the change's build, and review and verify follow at the depths its mode gives\n",{"name":5,"description":7},"main","c4a6bc27da835d109348b8708301821838e98db2","corpus\u002Fskills\u002Fsweep-tests","https:\u002F\u002Fgithub.com\u002Ffern-works\u002Fgreenline","https:\u002F\u002Fgithub.com\u002Ffern-works","2026-10-21 15:51:59",{"source":8,"skillName":5,"commands":31,"prompt":41,"zipAvailable":42},[32,35,38],{"tool":33,"command":34},"npx","npx skills add fern-works\u002Fgreenline --skill sweep-tests",{"tool":36,"command":37},"bunx","bunx skills add fern-works\u002Fgreenline --skill sweep-tests",{"tool":39,"command":40},"pnpm","pnpm dlx skills add fern-works\u002Fgreenline --skill sweep-tests","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Ffern-works\u002Fgreenline\u002Fblob\u002Fmain\u002Fcorpus\u002Fskills\u002Fsweep-tests\u002FSKILL.md",true,[44,50],{"path":45,"resourceType":46,"resourceTypeDesc":47,"sizeBytes":48,"sourceHtmlUrl":49},"bin\u002Fmutate.mjs","SCRIPT","脚本",9066,"https:\u002F\u002Fgithub.com\u002Ffern-works\u002Fgreenline\u002Fblob\u002Fmain\u002Fcorpus\u002Fskills\u002Fsweep-tests\u002Fbin\u002Fmutate.mjs",{"path":51,"resourceType":52,"resourceTypeDesc":53,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Fgreenline.md","REFERENCE","参考文档",357,"https:\u002F\u002Fgithub.com\u002Ffern-works\u002Fgreenline\u002Fblob\u002Fmain\u002Fcorpus\u002Fskills\u002Fsweep-tests\u002Freferences\u002Fgreenline.md",[57,64,69,76,83,90],{"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},"01M7D04ARSVW93Y1EGVM7HQQR8","use-garden","Consulting garden, the separately installed guidance command this repository's owner enabled, when an open engineering choice the repository does not settle, made in a change or only recommended, would be informed by published guidance. Use for a choice between approaches, a pattern or pitfall for a technology or concern the work touches, or a unit a repository decision names; not for a question about this repository's own facts or a change whose choices are already settled.","https:\u002F\u002Fgithub.com\u002Ffern-works\u002Fgreenline\u002Fblob\u002Fmain\u002Ftest\u002Fgoldens\u002Fgarden-enabled\u002Fdot-claude\u002Fskills\u002Fuse-garden\u002FSKILL.md",[],[],{"skillNo":65,"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":66,"categories":67,"keywords":68,"publishedAt":19},"01M7D045RXH0SEYZKHHW2YD0FQ","https:\u002F\u002Fgithub.com\u002Ffern-works\u002Fgreenline\u002Fblob\u002Fmain\u002Ftest\u002Fgoldens\u002Fgarden-enabled\u002Fdot-agents\u002Fskills\u002Fuse-garden\u002FSKILL.md",[],[],{"skillNo":70,"name":71,"nameZh":6,"description":72,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":19},"01M7D03ZXQHBP1RASSRHH5DX7S","what-to-build","Propose what this product could do next at three altitudes: sharpen a shipped slice, extend one, or open ground the product does not stand on yet. Use for 'what should we build next', 'any ideas for this app', 'what's missing', 'is this done or is there more in it', and quick-win hunts. Not for refactors, which belong to the architecture survey.","https:\u002F\u002Fgithub.com\u002Ffern-works\u002Fgreenline\u002Fblob\u002Fmain\u002Ftest\u002Fgoldens\u002Fcorpus\u002Fdot-claude\u002Fskills\u002Fwhat-to-build\u002FSKILL.md",[],[],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":19},"01M7D03WVYR3NN0N0CTN51RWBV","wayfinder","Plan a huge chunk of work (more than one agent session can hold) as a shared map of decision tickets in a plan directory's map.md, and resolve them one at a time until the way to the destination is clear.","https:\u002F\u002Fgithub.com\u002Ffern-works\u002Fgreenline\u002Fblob\u002Fmain\u002Ftest\u002Fgoldens\u002Fcorpus\u002Fdot-claude\u002Fskills\u002Fwayfinder\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":19},"01M7D03R6MAHHKQ42E5NBRHSF6","verify","Prove a change's claims with fresh evidence: restate each item of its acceptance so it can fail, capture a baseline and a treatment with the same command, compare the raw artifacts, and return VERIFIED, NOT VERIFIED or INCONCLUSIVE for each, at the depth the change's mode gives the verify step. Use when the mode reaches verify, or when the owner says \"verify this\", \"prove it works\" or \"did this fix it\".","https:\u002F\u002Fgithub.com\u002Ffern-works\u002Fgreenline\u002Fblob\u002Fmain\u002Ftest\u002Fgoldens\u002Fcorpus\u002Fdot-claude\u002Fskills\u002Fverify\u002FSKILL.md",[],[],{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"publishedAt":19},"01M7D03JYGW4945FSJV6CDE2K9","tdd","Test-first development at any step: discover the repository's own test commands, name the break a test catches, watch it fail for the intended reason, make it pass with the smallest change, and add a new test only where it catches a realistic failure the existing ones would miss. Use when writing or changing a test, building a feature or fixing a bug test-first, or deciding whether a change needs a new test; also on 'red-green-refactor' and integration tests.","https:\u002F\u002Fgithub.com\u002Ffern-works\u002Fgreenline\u002Fblob\u002Fmain\u002Ftest\u002Fgoldens\u002Fcorpus\u002Fdot-claude\u002Fskills\u002Ftdd\u002FSKILL.md",[],[]]