[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7AAHFCMZPQFAZ3ERM8A66AD":3,"skill-same:01M7AAHFCMZPQFAZ3ERM8A66AD":45},{"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":29,"updatedAt":30,"install":31,"resources":44},"01M7AAHFCMZPQFAZ3ERM8A66AD","work-on-ticket",null,"Work a rasp ticket end to end — read the Linear issue, check deps, move it to In Progress, branch, implement, verify every acceptance criterion, open a PR, and close the issue out. Use this whenever a milestone ID appears (M0-02, M1-09, M5-04, P2-SUBAGENT), or the user says \"next ticket\", \"pick up the CI one\", \"start M3-04\", \"implement the workspace confinement\", or otherwise asks for work that maps to a milestone item — even when they never say the word \"ticket\". Also use when finishing, reviewing, or closing one out.","theocod3s\u002Frasp","theocod3s","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F49566298?v=4","01M7AAH13XW4RE85V9JYH958Q7","Theophilus Okwugwuni","01M7AAH15040C5KR74SZM7PBR8",0,"MIT","https:\u002F\u002Fgithub.com\u002Ftheocod3s\u002Frasp\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fwork-on-ticket\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: work-on-ticket\ndescription: Work a rasp ticket end to end — read the Linear issue, check deps, move it to In Progress, branch, implement, verify every acceptance criterion, open a PR, and close the issue out. Use this whenever a milestone ID appears (M0-02, M1-09, M5-04, P2-SUBAGENT), or the user says \"next ticket\", \"pick up the CI one\", \"start M3-04\", \"implement the workspace confinement\", or otherwise asks for work that maps to a milestone item — even when they never say the word \"ticket\". Also use when finishing, reviewing, or closing one out.\n---\n\n# Working a rasp ticket\n\n~91 MVP items. Doing them consistently matters more than doing any one cleverly. This is the\nsequence, and the reasoning behind the parts that are easy to skip.\n\n## 1. Read the ticket, and its spec\n\nTickets live in Linear and nowhere else — [project Rasp](https:\u002F\u002Flinear.app\u002Ftheocod3s\u002Fproject\u002Frasp-be0653f32d76),\none milestone per M0–M5 plus one per future phase. Find one by its ID: search the project for\n`M0-02` and take the exact title match (`M0-02 · CI`); fuzzy matches rank below it.\n\nEvery ticket names a **spec anchor** (`design §13`, `prd §6.6`). Read that section before\nwriting anything: the ticket is a decomposition, not the requirement, and where the two\ndisagree the spec wins.\n\nCheck **Depends on**. They live in the description as prose, not as Linear blocking relations,\nso nothing enforces them and nothing will stop you starting a ticket whose dependency is\nunmerged. If one is, say so and stop rather than working around it.\n\n**Don't invent scope.** If something looks missing it is usually deliberate — check\n`docs\u002Fscope.md`'s \"deliberately excluded\" before adding it. If it's a real gap, raise it; don't\nsilently fill it. `VISION.md`'s closing accept\u002Fresist tests are the acceptance policy: a change\nthat passes the ticket but fails them is a conflict to raise, not to ship.\n\n## 2. Claim it, then branch\n\n**Move the issue to In Progress before writing anything**, even for a ticket that will take ten\nminutes. Linear is the only record of where the work stands, so a ticket in Todo while a branch\nexists is the same failure as one left In Progress after merge (§6): the next session reads the\nboard and gets the wrong answer.\n\nBranch `rasp\u002F\u003Cid-lowercased>-\u003Cslug>` — `rasp\u002Fm0-02-ci` — from an up-to-date `main`.\n\n## 3. Implement\n\nRead `AGENTS.md` first if you haven't this session. It carries the invariants that are cheapest\nto break by accident, and violating one is far more expensive than the ticket is worth.\n\n**Comments: follow `.agents\u002Fskills\u002Fcomment-density\u002FSKILL.md` while writing, not after.** The\nhabit it stops is invisible from inside — a review round asks \"why?\", the answer is appended to\nthe comment, nothing is ever taken back out, every round is an improvement and the file still\nends up a third prose. M0-06 shipped `internal\u002Fllm` at 45% comment lines that way. **The default\nis bare code.** An exported identifier owes `go doc` one sentence; an unexported one owes\nnothing; no function gets a comment for merely existing. Past that, a comment earns its place\nonly by recording what the code **cannot** say — why a rule exists, which provider wire shape\nforced it, why it is deliberately *not* stricter — and says it in a sentence or two, not a\nparagraph. Cut what restates the identifier, and cut what the error string three lines below\nalready says in full sentences.\n\n## 4. Verify every acceptance criterion, one at a time\n\nThe criteria are the definition of done, written so someone who didn't do the work can check\nthem. Go through them individually rather than concluding \"it works\".\n\n**A check you have only seen pass is not evidence.** If the ticket adds a test, a lint, a CI\ngate — anything whose job is to fail — break the thing it guards on purpose and watch it fail,\nthen restore. This is the single highest-value habit in this repo: it is how all three of the\nbugs behind AGENTS.md's *\"a check that cannot run must fail, not pass\"* were caught, and none\nof them was visible from a green run.\n\nThe trap inside that habit: **a break that stops the package compiling proves nothing.** M0-04\nreplaced the string-aware JSONC comment stripper with a naive one and the suite went red — but\nred from the build, so no test ran and none was shown to catch anything. `if false {` does it\ntoo, whenever it leaves a variable or an import unused. If your break produces a build error, it\nis the compiler that caught you, not the check.\n\nAnd **assert the right failure, not just a failure.** M0-05's timeout test checked that *some*\nerror came back. A later fix broke working credential helpers by returning a different error a\nsecond sooner, and the test stayed green — passing precisely when the thing it guarded was being\ndestroyed. `err != nil` is the assertion most likely to survive its own subject.\n\n**Mutate what a test rests on, not only what it asserts.** The check being weak is the failure\nyou look for; the test being incapable of failing is the one you find. M0-07 shipped three at\nonce — `defer stream.Close()` could be deleted with the whole suite green, a usage check read\nthe same pointer before and after the drain, and a third pinned the SDK's JSON key order, which\nwould go quiet the day that order changed. All three read fine, and all three surfaced only when\nsomething they silently rested on was broken. So mutate the setup line, the helper and the\nliteral nothing asserts on — and make a helper that finds nothing to examine say so, because\nexamining nothing is the quietest pass there is.\n\nOnce there is more than a handful, script it: apply one mutation, run the tests, restore, print\ncaught\u002Fnot-caught per case. M0-04 ran twenty-four that way in under a minute, which is the\ndifference between doing this for the two obvious guards and doing it for all of them. **Restore\nby writing back the bytes you replaced** — never `git checkout -- .`, which resets the whole tree\nto HEAD and takes every uncommitted change with it. It did.\n\n**A not-caught result has three causes and only one is the check's fault.** The check may be\nweak; the test may never reach the line you mutated; or the `-run` pattern may match no test at\nall, which exits 0 and reads exactly like a pass. M0-06 hit the second (a guard reached only\nafter the stream had already been abandoned) and the third (a subtest name with an apostrophe in\nit). Make the harness shout when nothing ran, and confirm the line runs, before touching the\ncheck.\n\nThen `just ci` — fmt-check, vet, build, test, race. Run it before pushing, not after.\n\n## 5. Open the PR\n\n**Write the description for an engineering director.** Someone two levels out from the code\nshould read it and know what shipped and why it matters — so lead with what the PR *does*, in\nplain language, then only what the reviewer genuinely needs: a judgement call and its\nalternative, a risk, a piece deliberately left out. **Don't list the acceptance criteria**; they\nlive in the ticket, and copying them turns the description into a form nobody reads.\n\n**Keep it short enough that the reader reaches the diff.** M0-04's body ran past a thousand words\nre-arguing reasoning already in the commits and the code — and a description nobody finishes\nfails at its only job. If a decision needs a paragraph to defend, that paragraph belongs in the\ncode, where the next reader is actually standing.\n\nDon't merge your own PR unmarked. Run `\u002Fcode-review`, and treat its findings as claims to check\nrather than facts — it has a real false-positive rate, and rejecting a wrong finding with\nreasoning is as valuable as fixing a right one.\n\n**Then remember the fixes are code nobody has reviewed**, so any pass whose fixes changed\nbehaviour earns another. M0-05 took five passes for 17 findings, and pass 3's fix *caused* pass\n4's worst bug: `cmd.WaitDelay` starts its timer when the child exits, so the fix for a helper that\nhung became a failure for every helper that worked.\n\n**Stop at the first pass that finds no defect in behaviour** — fix its wording findings, but don't\nre-run for them. Waiting instead for a pass that finds *nothing*, or that reports its own last\nfindings already fixed, does not terminate: M0-06a ran nine passes and never got one, because\nnaming and phrasing findings do not run out. Its substance stopped moving at pass three.\n\n**Record anything the work settled that outlives it.** If the ticket fixed a rule new code could\nquietly reverse — one every adapter must follow, a format nobody may change, a tempting fix that\nis wrong — add it to `docs\u002Fdecisions.md` in this PR, while the reasoning is still in your head.\n\nThe bar is high on purpose, and it is not \"what did I choose\". It is: would someone who never saw\nthis work re-derive it, or undo it believing they were fixing something? Three things fail that\nbar. What is **still open** belongs in the Linear issue that will settle it, not here. What shapes\n**one file** belongs in that file, where the reader is standing. What a **document already says**\nstays there — `design.md` is the primary reference and this does not restate it. Most tickets add\nnothing, and that is the expected outcome.\n\n## 6. Merge and close out\n\nSquash-merge, with a message you wrote rather than the one GitHub concatenates — see the rule\nin `AGENTS.md`. Then move the issue to Done in the same pass — the other half of §2's rule, and\nan issue left In Progress after merge misleads exactly as much as one never started.\n\nIf the ticket's shape changed while you worked it, edit the issue description to match what was\nactually built. A ticket that no longer describes its own outcome is worse than no ticket,\nbecause it still reads as authoritative.\n\n## 7. Report back\n\nOnce the PR is merged and the issue is Done, **write the closing summary for a CTO**: someone who\nknows the project but has not seen this work, will not open the diff, and cannot be assumed to\nknow what any identifier or section number refers to. So expand a term the first time it appears,\nand describe every change by what it now *does* before naming what it touched.\n\n- One or two sentences on what the work delivers, in plain language.\n- Then **file by file, what changed in each** — a sentence or two per file, its new behaviour and\n  the reason for it, not a restatement of the diff.\n- Anything that genuinely needs a decision.\n\nShort, not thin: no reasoning trail, no verification narrative, no caveats gathered along the\nway — detail earns its place *before* the work, or when something failed or went differently than\nasked. But short is not the same as understandable: a wall of identifiers standing in for\nsentences is a report that has to be decoded, which is a report not delivered.\n\nThis is the only place the closing summary is specified. `AGENTS.md` carried a second version\nuntil the two asked for opposite reports; don't start a third.\n\n---\n\n## Revising this skill\n\nThis file should get better as tickets teach us things. After closing one out, ask: **did\nanything here mislead me, go missing, or have to be re-derived?** If so, amend it in the\nticket's own PR, so the change is reviewed alongside the work that motivated it.\n\nTwo things keep this from rotting into a changelog:\n\n**Only record what changes what someone does next time.** A rule that applies to one ticket is\na note in that PR, not a line here. If you can't name a future ticket it would have helped,\nleave it out.\n\n**Prune as you add.** Target ~150 lines. When you add something, look for what has stopped\nearning its place — a step now enforced by a test, advice that turned out situational, a caveat\nabout a milestone long since shipped. A skill nobody reads to the end fails exactly the way a\n2,100-line design document does, which is why this file exists at all.\n\nSay *why*, not just what. \"Verify checks by breaking them\" is forgettable; the `fmt-check`\nexample is not.\n","# Working a rasp ticket\n\n~91 MVP items. Doing them consistently matters more than doing any one cleverly. This is the\nsequence, and the reasoning behind the parts that are easy to skip.\n\n## 1. Read the ticket, and its spec\n\nTickets live in Linear and nowhere else — [project Rasp](https:\u002F\u002Flinear.app\u002Ftheocod3s\u002Fproject\u002Frasp-be0653f32d76),\none milestone per M0–M5 plus one per future phase. Find one by its ID: search the project for\n`M0-02` and take the exact title match (`M0-02 · CI`); fuzzy matches rank below it.\n\nEvery ticket names a **spec anchor** (`design §13`, `prd §6.6`). Read that section before\nwriting anything: the ticket is a decomposition, not the requirement, and where the two\ndisagree the spec wins.\n\nCheck **Depends on**. They live in the description as prose, not as Linear blocking relations,\nso nothing enforces them and nothing will stop you starting a ticket whose dependency is\nunmerged. If one is, say so and stop rather than working around it.\n\n**Don't invent scope.** If something looks missing it is usually deliberate — check\n`docs\u002Fscope.md`'s \"deliberately excluded\" before adding it. If it's a real gap, raise it; don't\nsilently fill it. `VISION.md`'s closing accept\u002Fresist tests are the acceptance policy: a change\nthat passes the ticket but fails them is a conflict to raise, not to ship.\n\n## 2. Claim it, then branch\n\n**Move the issue to In Progress before writing anything**, even for a ticket that will take ten\nminutes. Linear is the only record of where the work stands, so a ticket in Todo while a branch\nexists is the same failure as one left In Progress after merge (§6): the next session reads the\nboard and gets the wrong answer.\n\nBranch `rasp\u002F\u003Cid-lowercased>-\u003Cslug>` — `rasp\u002Fm0-02-ci` — from an up-to-date `main`.\n\n## 3. Implement\n\nRead `AGENTS.md` first if you haven't this session. It carries the invariants that are cheapest\nto break by accident, and violating one is far more expensive than the ticket is worth.\n\n**Comments: follow `.agents\u002Fskills\u002Fcomment-density\u002FSKILL.md` while writing, not after.** The\nhabit it stops is invisible from inside — a review round asks \"why?\", the answer is appended to\nthe comment, nothing is ever taken back out, every round is an improvement and the file still\nends up a third prose. M0-06 shipped `internal\u002Fllm` at 45% comment lines that way. **The default\nis bare code.** An exported identifier owes `go doc` one sentence; an unexported one owes\nnothing; no function gets a comment for merely existing. Past that, a comment earns its place\nonly by recording what the code **cannot** say — why a rule exists, which provider wire shape\nforced it, why it is deliberately *not* stricter — and says it in a sentence or two, not a\nparagraph. Cut what restates the identifier, and cut what the error string three lines below\nalready says in full sentences.\n\n## 4. Verify every acceptance criterion, one at a time\n\nThe criteria are the definition of done, written so someone who didn't do the work can check\nthem. Go through them individually rather than concluding \"it works\".\n\n**A check you have only seen pass is not evidence.** If the ticket adds a test, a lint, a CI\ngate — anything whose job is to fail — break the thing it guards on purpose and watch it fail,\nthen restore. This is the single highest-value habit in this repo: it is how all three of the\nbugs behind AGENTS.md's *\"a check that cannot run must fail, not pass\"* were caught, and none\nof them was visible from a green run.\n\nThe trap inside that habit: **a break that stops the package compiling proves nothing.** M0-04\nreplaced the string-aware JSONC comment stripper with a naive one and the suite went red — but\nred from the build, so no test ran and none was shown to catch anything. `if false {` does it\ntoo, whenever it leaves a variable or an import unused. If your break produces a build error, it\nis the compiler that caught you, not the check.\n\nAnd **assert the right failure, not just a failure.** M0-05's timeout test checked that *some*\nerror came back. A later fix broke working credential helpers by returning a different error a\nsecond sooner, and the test stayed green — passing precisely when the thing it guarded was being\ndestroyed. `err != nil` is the assertion most likely to survive its own subject.\n\n**Mutate what a test rests on, not only what it asserts.** The check being weak is the failure\nyou look for; the test being incapable of failing is the one you find. M0-07 shipped three at\nonce — `defer stream.Close()` could be deleted with the whole suite green, a usage check read\nthe same pointer before and after the drain, and a third pinned the SDK's JSON key order, which\nwould go quiet the day that order changed. All three read fine, and all three surfaced only when\nsomething they silently rested on was broken. So mutate the setup line, the helper and the\nliteral nothing asserts on — and make a helper that finds nothing to examine say so, because\nexamining nothing is the quietest pass there is.\n\nOnce there is more than a handful, script it: apply one mutation, run the tests, restore, print\ncaught\u002Fnot-caught per case. M0-04 ran twenty-four that way in under a minute, which is the\ndifference between doing this for the two obvious guards and doing it for all of them. **Restore\nby writing back the bytes you replaced** — never `git checkout -- .`, which resets the whole tree\nto HEAD and takes every uncommitted change with it. It did.\n\n**A not-caught result has three causes and only one is the check's fault.** The check may be\nweak; the test may never reach the line you mutated; or the `-run` pattern may match no test at\nall, which exits 0 and reads exactly like a pass. M0-06 hit the second (a guard reached only\nafter the stream had already been abandoned) and the third (a subtest name with an apostrophe in\nit). Make the harness shout when nothing ran, and confirm the line runs, before touching the\ncheck.\n\nThen `just ci` — fmt-check, vet, build, test, race. Run it before pushing, not after.\n\n## 5. Open the PR\n\n**Write the description for an engineering director.** Someone two levels out from the code\nshould read it and know what shipped and why it matters — so lead with what the PR *does*, in\nplain language, then only what the reviewer genuinely needs: a judgement call and its\nalternative, a risk, a piece deliberately left out. **Don't list the acceptance criteria**; they\nlive in the ticket, and copying them turns the description into a form nobody reads.\n\n**Keep it short enough that the reader reaches the diff.** M0-04's body ran past a thousand words\nre-arguing reasoning already in the commits and the code — and a description nobody finishes\nfails at its only job. If a decision needs a paragraph to defend, that paragraph belongs in the\ncode, where the next reader is actually standing.\n\nDon't merge your own PR unmarked. Run `\u002Fcode-review`, and treat its findings as claims to check\nrather than facts — it has a real false-positive rate, and rejecting a wrong finding with\nreasoning is as valuable as fixing a right one.\n\n**Then remember the fixes are code nobody has reviewed**, so any pass whose fixes changed\nbehaviour earns another. M0-05 took five passes for 17 findings, and pass 3's fix *caused* pass\n4's worst bug: `cmd.WaitDelay` starts its timer when the child exits, so the fix for a helper that\nhung became a failure for every helper that worked.\n\n**Stop at the first pass that finds no defect in behaviour** — fix its wording findings, but don't\nre-run for them. Waiting instead for a pass that finds *nothing*, or that reports its own last\nfindings already fixed, does not terminate: M0-06a ran nine passes and never got one, because\nnaming and phrasing findings do not run out. Its substance stopped moving at pass three.\n\n**Record anything the work settled that outlives it.** If the ticket fixed a rule new code could\nquietly reverse — one every adapter must follow, a format nobody may change, a tempting fix that\nis wrong — add it to `docs\u002Fdecisions.md` in this PR, while the reasoning is still in your head.\n\nThe bar is high on purpose, and it is not \"what did I choose\". It is: would someone who never saw\nthis work re-derive it, or undo it believing they were fixing something? Three things fail that\nbar. What is **still open** belongs in the Linear issue that will settle it, not here. What shapes\n**one file** belongs in that file, where the reader is standing. What a **document already says**\nstays there — `design.md` is the primary reference and this does not restate it. Most tickets add\nnothing, and that is the expected outcome.\n\n## 6. Merge and close out\n\nSquash-merge, with a message you wrote rather than the one GitHub concatenates — see the rule\nin `AGENTS.md`. Then move the issue to Done in the same pass — the other half of §2's rule, and\nan issue left In Progress after merge misleads exactly as much as one never started.\n\nIf the ticket's shape changed while you worked it, edit the issue description to match what was\nactually built. A ticket that no longer describes its own outcome is worse than no ticket,\nbecause it still reads as authoritative.\n\n## 7. Report back\n\nOnce the PR is merged and the issue is Done, **write the closing summary for a CTO**: someone who\nknows the project but has not seen this work, will not open the diff, and cannot be assumed to\nknow what any identifier or section number refers to. So expand a term the first time it appears,\nand describe every change by what it now *does* before naming what it touched.\n\n- One or two sentences on what the work delivers, in plain language.\n- Then **file by file, what changed in each** — a sentence or two per file, its new behaviour and\n  the reason for it, not a restatement of the diff.\n- Anything that genuinely needs a decision.\n\nShort, not thin: no reasoning trail, no verification narrative, no caveats gathered along the\nway — detail earns its place *before* the work, or when something failed or went differently than\nasked. But short is not the same as understandable: a wall of identifiers standing in for\nsentences is a report that has to be decoded, which is a report not delivered.\n\nThis is the only place the closing summary is specified. `AGENTS.md` carried a second version\nuntil the two asked for opposite reports; don't start a third.\n\n---\n\n## Revising this skill\n\nThis file should get better as tickets teach us things. After closing one out, ask: **did\nanything here mislead me, go missing, or have to be re-derived?** If so, amend it in the\nticket's own PR, so the change is reviewed alongside the work that motivated it.\n\nTwo things keep this from rotting into a changelog:\n\n**Only record what changes what someone does next time.** A rule that applies to one ticket is\na note in that PR, not a line here. If you can't name a future ticket it would have helped,\nleave it out.\n\n**Prune as you add.** Target ~150 lines. When you add something, look for what has stopped\nearning its place — a step now enforced by a test, advice that turned out situational, a caveat\nabout a milestone long since shipped. A skill nobody reads to the end fails exactly the way a\n2,100-line design document does, which is why this file exists at all.\n\nSay *why*, not just what. \"Verify checks by breaking them\" is forgettable; the `fmt-check`\nexample is not.\n",{"name":5,"description":7},"main","f0c4fdf29d31460de17b211bd29b3494cbba28ad",".agents\u002Fskills\u002Fwork-on-ticket","https:\u002F\u002Fgithub.com\u002Ftheocod3s\u002Frasp","https:\u002F\u002Fgithub.com\u002Ftheocod3s","AI Engineer. Software Engineer 👨🏾‍💻.","2026-09-04 08:40:48",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add theocod3s\u002Frasp --skill work-on-ticket",{"tool":37,"command":38},"bunx","bunx skills add theocod3s\u002Frasp --skill work-on-ticket",{"tool":40,"command":41},"pnpm","pnpm dlx skills add theocod3s\u002Frasp --skill work-on-ticket","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Ftheocod3s\u002Frasp\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fwork-on-ticket\u002FSKILL.md",true,[],[46,51,57,63,69],{"skillNo":47,"name":5,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":48,"categories":49,"keywords":50,"publishedAt":19},"01M7AAHVK7NZRZG0THJDAVMPPW","https:\u002F\u002Fgithub.com\u002Ftheocod3s\u002Frasp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fwork-on-ticket\u002FSKILL.md",[],[],{"skillNo":52,"name":53,"nameZh":6,"description":6,"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},"01M7AAHS15A2ZRDBA97WYWC72X","work-on-milestone","https:\u002F\u002Fgithub.com\u002Ftheocod3s\u002Frasp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fwork-on-milestone\u002FSKILL.md",[],[],{"skillNo":58,"name":59,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":19},"01M7AAHP3ZWF3ERE2A9X0QP3Z3","comment-density","https:\u002F\u002Fgithub.com\u002Ftheocod3s\u002Frasp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fcomment-density\u002FSKILL.md",[],[],{"skillNo":64,"name":53,"nameZh":6,"description":65,"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},"01M7AAH99FXKBMVS2T78W2HKPP","Use when the ask spans more than one milestone item — \"work M1\", \"run the milestone\", \"finish the rest of the milestone\", \"do the next three tickets\", \"keep going until blocked\". A single milestone ID (\"implement M1-25\") is work-on-ticket's trigger, not this one.","https:\u002F\u002Fgithub.com\u002Ftheocod3s\u002Frasp\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fwork-on-milestone\u002FSKILL.md",[],[],{"skillNo":70,"name":59,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":19},"01M7AAH34Q41YKD1REJJMFQ7TZ","Decide which comments to write, keep, compress or cut in rasp's Go code. Use when writing a new file or package, when a review says a comment is unclear or unnecessary, when a file's comment-to-code ratio is climbing past ~25%, and as the last pass before opening a PR. Also use when asked to \"clean up the comments\", \"this is over-commented\", \"trim the docs in \u003Cpackage>\", or when a review loop has been adding justification to comments as it goes.","https:\u002F\u002Fgithub.com\u002Ftheocod3s\u002Frasp\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fcomment-density\u002FSKILL.md",[],[]]