[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7AW4T0VTDCJ17CKCSYCV6VR":3,"skill-same:01M7AW4T0VTDCJ17CKCSYCV6VR":52},{"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":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":6,"ownerDescription":30,"updatedAt":31,"install":32,"resources":45},"01M7AW4T0VTDCJ17CKCSYCV6VR","yun-diagnose-bug",null,"Diagnosis loop for hard bugs and performance regressions — build a tight feedback loop that goes red on the bug before theorising about it. Use when the user says \"diagnose this bug\" \u002F \"diagnosticá\" \u002F \"no entiendo por qué falla\" \u002F \"está lento\", when a bug resisted the obvious fix, when its repro will not hold still, or when something got slower.","matiinchauspe\u002Fyunque","matiinchauspe","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F7861346?v=4","01M7AW106HSV1DVWJSQQBD6BYB","Matias","01M7AW107JZVCTQHZEED16N5PZ",1,"MIT","https:\u002F\u002Fgithub.com\u002Fmatiinchauspe\u002Fyunque\u002Fblob\u002Fmain\u002Fskills\u002Fyun-diagnose-bug\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: yun-diagnose-bug\ndescription: Diagnosis loop for hard bugs and performance regressions — build a tight feedback loop that goes red on the bug before theorising about it. Use when the user says \"diagnose this bug\" \u002F \"diagnosticá\" \u002F \"no entiendo por qué falla\" \u002F \"está lento\", when a bug resisted the obvious fix, when its repro will not hold still, or when something got slower.\n---\n\n# yun-diagnose-bug\n\nA discipline for hard bugs. The whole skill rests on one move: get a **tight** feedback loop\nthat goes **red** on this bug before you theorise about it. Skip a step only with an explicit\nreason, and never step 1.\n\nBefore you start, read what the project already knows. **Recall** prior diagnoses of this\nsymptom through `skills\u002F_shared\u002Fmemory-convention.md` — a past cause and the loop that caught\nit is a head start on step 3's ranking, never a licence to skip building the loop. And\n**consult** the project's domain model through `skills\u002F_shared\u002Fdomain-convention.md` so you\nname modules and behaviour in its **ubiquitous language**, and respect the ADRs in the area\nyou're touching. And **conform** to the repo's practice through\n`skills\u002F_shared\u002Fpractice-convention.md` so the failing test you write and the fix you land are\nshaped the way this repo already shapes them. All three degrade silently when there is nothing\nto read.\n\n## Redact\n\nThis skill has you show commands, outputs and captured artifacts. **Redact every secret\nfirst**: write `\u003CREDACTED>` in its place. Build loops against env vars so the credential stays\nin the environment rather than in what you show, and quote only the lines of a captured\nartifact that carry the signal. If what survives redaction is not enough to diagnose the bug,\nsay so and ask the user.\n\nCaptured artifacts — traces, HAR files, log dumps, core dumps — carry live secrets whatever you\nquote from them. Write them to a **gitignored scratch path** outside the working tree, and\ndelete them in step 6: the commit that lands the fix must not be able to reach one.\n\n## Steps\n\n### 1. Build a feedback loop\n\n**This is the skill.** With a **tight** pass\u002Ffail signal — one that goes **red** on _this_\nbug — you will find the cause; bisection, hypothesis-testing and instrumentation all just\nconsume it. Spend disproportionate effort here, and be **relentless**.\n\n**Pick a route to construct one.** [`LOOPS.md`](LOOPS.md) holds the eleven, ordered — read it\nand pick before you build, because the route sets the ceiling on how **tight** the loop can get.\nTwo of them carry this skill's tails: the **measurement harness**, which records a **baseline**\nand goes **red** past a threshold, is the only route a performance regression has, and the\n**human-in-the-loop** script is the last resort when no route runs unattended.\n\n**Tighten it.** Treat the loop as a product: faster (cache setup, skip unrelated init, narrow\nthe scope), sharper (assert the specific symptom, not \"didn't crash\"), more deterministic (pin\ntime, seed RNG, isolate the filesystem, freeze the network). A 30-second flaky loop is barely\nbetter than none; a 2-second deterministic one is a superpower.\n\n**When it won't reproduce cleanly.** The goal is not a clean repro but a **higher reproduction\nrate**. Loop the trigger 100×, parallelise, add stress, narrow timing windows, inject sleeps. A\n50%-flake bug is debuggable; a 1% one is not — keep raising the rate until it is.\n\n**When it won't reproduce at all.** Stop and say so. List what you tried, then ask the user for\none of: access to an environment that reproduces it; a redacted captured artifact — HAR file,\nlog dump, core dump, timestamped screen recording; or permission to add temporary production\ninstrumentation. Wait for one of those rather than reasoning your way to a cause you cannot\nobserve.\n\n**No loop, no step 2.** If you catch yourself reading code to build a theory before that\ncommand exists, stop — jumping straight to a hypothesis is the exact failure this skill exists\nto prevent.\n\nDone when you can name **one command** you have already run at least once — showing the\ninvocation and its redacted output — that is:\n\n- [ ] **Red-capable** — it drives the real bug code path and asserts the user's exact symptom,\n      so it goes red on this bug and green once fixed. Not \"runs without erroring\". For a\n      regression, red is a measurement breaching its threshold against the baseline.\n- [ ] **Deterministic** — the same verdict every run (flaky bugs: a pinned, high reproduction\n      rate).\n- [ ] **Fast** — seconds, not minutes.\n- [ ] **Agent-runnable** — runnable unattended; a human inside it only via the human-in-the-loop\n      script.\n\nOr, when it will not reproduce at all, done when you have stopped and asked — attempts listed,\none of the three asks above made.\n\n### 2. Reproduce and minimise\n\nRun the loop and watch it go **red**. Confirm it produces the failure the **user** described\nrather than a different one nearby — wrong bug, wrong fix — that it reproduces across runs, and\nthat you have captured the exact symptom for later steps to verify a fix against.\n\nThen shrink the repro to the smallest scenario that still goes red: cut inputs, callers,\nconfig, data and steps **one at a time**, re-running the loop after each cut. Keep the\nun-minimised loop as you go — step 5 re-runs it. It pays twice: fewer moving parts left to\nsuspect in step 3, and the clean regression test in step 5.\n\nOn a flaky loop, one green run after a cut is luck rather than evidence. Re-run at the\nreproduction rate you pinned in step 1 — enough runs that a still-load-bearing element would\nhave gone red — before you accept the cut.\n\n**Performance branch.** Scale is itself part of the signal: cutting data volume, iteration\ncount or concurrency takes the measurement harness under its threshold without removing a\ncause. Pin the scale that reproduces the regression, minimise around it, and treat that pinned\nscale as load-bearing.\n\nDone when the loop has gone red on the **user's** symptom across runs with that symptom\ncaptured, the un-minimised loop is still runnable, and every remaining element is load-bearing:\nremoving any one of them turns the loop green.\n\n### 3. Hypothesise\n\nGenerate **3–5 ranked hypotheses** before testing any of them — generating one at a time\nanchors you on the first plausible idea. Each must be falsifiable, stating its prediction: \"if\nX is the cause, then changing Y makes the bug disappear \u002F changing Z makes it worse\". A\nhypothesis with no prediction is a vibe: sharpen it or drop it.\n\nShow the ranked list to the user before testing. They re-rank it instantly on domain knowledge\n(\"we just deployed a change to #3\") or name the ones they have already ruled out. Don't block\non it — proceed with your own ranking if the user is away.\n\nDone when the ranked, falsifiable list exists and has been shown.\n\n### 4. Instrument\n\nEach probe maps to a specific prediction from step 3, and you **change one variable at a\ntime**. Reach for a debugger or REPL first where the environment supports one — one breakpoint\nbeats ten logs — then targeted logs at the boundaries that distinguish the hypotheses.\n\n**Tag every debug log** with a unique prefix, e.g. `[DEBUG-a4f2]`, so cleanup is a single\nsearch. Untagged logs survive; tagged logs die.\n\n**Performance branch.** For regressions, logs are usually the wrong instrument: measure against\nstep 1's baseline, then bisect. Measure first, fix second.\n\nDone when a probe has confirmed one hypothesis, or falsified all of them and sent you back to\nstep 3 with what it taught you. Three round trips without a confirmed hypothesis is the signal\nto stop, report what you have ruled out, and clean up through step 6.\n\n### 5. Fix, test-first\n\nWrite the regression test **before** the fix, but only where a **correct seam** exists: one at\nwhich the test exercises the real bug pattern as it occurs at the call site. A seam too shallow\nto replicate the chain that triggered the bug gives false confidence.\n\n**No correct seam is itself the finding.** Report it — the codebase's shape is what keeps this\nbug from being locked down. Still fix it, and say plainly that the lockdown is missing and why.\n\nWhere a correct seam does exist, **name it as the candidate and carry it into `yun-tdd`** — the\nminimised repro shows where the bug's chain actually runs, so the seam arrives with its\nevidence rather than being re-derived. A candidate is not an agreement: `yun-tdd`'s rule\nstands, and **the user** confirms the seam there before any test is written. Unlike step 3's\nranking, this gate **blocks** — a test at an unconfirmed seam is the false confidence this step\nopened by warning about. Then turn the minimised repro into a failing test at the confirmed\nseam and drive the fix through `yun-tdd`'s **red → green** loop, which owns the seam discipline\nand what makes a test worth keeping. Finally re-run step 1's loop against the original,\nun-minimised scenario.\n\n**Performance branch.** A timing assertion at a seam is a flaky test, and `yun-tdd`'s\n**red → green** loop has no baseline or threshold to drive. Step 1's measurement harness is\nalready the regression check: fix, then run it against the recorded baseline until it goes\ngreen under the threshold, and report the before and after numbers.\n\nDone when one of:\n\n- **Seam confirmed** — the regression test is written at it, it passes, and step 1's loop is\n  green against the un-minimised scenario.\n- **No correct seam** — its absence is reported with the reason the codebase's shape blocks the\n  lockdown, and step 1's loop is green against the un-minimised scenario without one.\n- **Performance regression** — step 1's measurement harness is green under its threshold against\n  the recorded baseline, with the before and after numbers reported.\n- **Seam unconfirmed, user away** — nothing is written at it. Park: report the confirmed cause\n  and the seam candidate, and leave the fix for the confirmation. The gate holds; the run ends\n  rather than stalling behind it.\n\n### 6. Clean up\n\n- [ ] Every `[DEBUG-...]` probe is removed — search the prefix.\n- [ ] Throwaway harnesses and any diagnostic test step 1 wrote are deleted — or, when the run\n      ends without a fix, moved to scratch and their paths reported, so the run that resumes\n      inherits the loop instead of rebuilding it.\n- [ ] Every captured artifact is deleted from scratch.\n- [ ] The hypothesis that proved correct — or, when step 4 aborted, the ones it ruled out — is\n      `persist`ed through `skills\u002F_shared\u002Fmemory-convention.md` with the symptom and the loop\n      that caught it, so the next run's recall starts where this one ended.\n\nThe fix lands where an attended run lands anywhere in this harness: commit it only when the\nuser asks.\n\nDone when every box is checked.\n\n## Attribution\n\nAdapted for this workspace from **Matt Pocock's `diagnosing-bugs`**\n(github.com\u002Fmattpocock\u002Fskills, MIT). Changes: renamed to the `yun-\u003Cverb>-\u003Cnoun>` convention; the\nphases become numbered steps, each closing on a **Done when** criterion; reading `GLOSSARY.md` and\nADRs becomes a `consult` through `skills\u002F_shared\u002Fdomain-convention.md`, and the correct hypothesis\na `persist` through `skills\u002F_shared\u002Fmemory-convention.md`, with a `recall` at the head to read it\nback. The write-test-watch-it-fail-apply-fix loop is handed to `yun-tdd`, the single source of\ntruth for seams and test quality. A measurement route is added to give the performance branch a\npath through step 1, and the routes are disclosed to [`LOOPS.md`](LOOPS.md). Named tools, the\n`hitl-loop` template and the `agents\u002Fopenai.yaml` wiring are dropped, since the harness runs\nagainst repos whose stack it does not choose. Upstream's \"never log everything and grep\" is\nrestated positively per this harness's negation rule, and Spanish triggers are added to the\ndescription. The loop-first discipline, the **tight** and **red** leading words, minimisation,\nranked falsifiable hypotheses, tagged probes, the performance branch and the cleanup checklist\nare preserved.\n","# yun-diagnose-bug\n\nA discipline for hard bugs. The whole skill rests on one move: get a **tight** feedback loop\nthat goes **red** on this bug before you theorise about it. Skip a step only with an explicit\nreason, and never step 1.\n\nBefore you start, read what the project already knows. **Recall** prior diagnoses of this\nsymptom through `skills\u002F_shared\u002Fmemory-convention.md` — a past cause and the loop that caught\nit is a head start on step 3's ranking, never a licence to skip building the loop. And\n**consult** the project's domain model through `skills\u002F_shared\u002Fdomain-convention.md` so you\nname modules and behaviour in its **ubiquitous language**, and respect the ADRs in the area\nyou're touching. And **conform** to the repo's practice through\n`skills\u002F_shared\u002Fpractice-convention.md` so the failing test you write and the fix you land are\nshaped the way this repo already shapes them. All three degrade silently when there is nothing\nto read.\n\n## Redact\n\nThis skill has you show commands, outputs and captured artifacts. **Redact every secret\nfirst**: write `\u003CREDACTED>` in its place. Build loops against env vars so the credential stays\nin the environment rather than in what you show, and quote only the lines of a captured\nartifact that carry the signal. If what survives redaction is not enough to diagnose the bug,\nsay so and ask the user.\n\nCaptured artifacts — traces, HAR files, log dumps, core dumps — carry live secrets whatever you\nquote from them. Write them to a **gitignored scratch path** outside the working tree, and\ndelete them in step 6: the commit that lands the fix must not be able to reach one.\n\n## Steps\n\n### 1. Build a feedback loop\n\n**This is the skill.** With a **tight** pass\u002Ffail signal — one that goes **red** on _this_\nbug — you will find the cause; bisection, hypothesis-testing and instrumentation all just\nconsume it. Spend disproportionate effort here, and be **relentless**.\n\n**Pick a route to construct one.** [`LOOPS.md`](LOOPS.md) holds the eleven, ordered — read it\nand pick before you build, because the route sets the ceiling on how **tight** the loop can get.\nTwo of them carry this skill's tails: the **measurement harness**, which records a **baseline**\nand goes **red** past a threshold, is the only route a performance regression has, and the\n**human-in-the-loop** script is the last resort when no route runs unattended.\n\n**Tighten it.** Treat the loop as a product: faster (cache setup, skip unrelated init, narrow\nthe scope), sharper (assert the specific symptom, not \"didn't crash\"), more deterministic (pin\ntime, seed RNG, isolate the filesystem, freeze the network). A 30-second flaky loop is barely\nbetter than none; a 2-second deterministic one is a superpower.\n\n**When it won't reproduce cleanly.** The goal is not a clean repro but a **higher reproduction\nrate**. Loop the trigger 100×, parallelise, add stress, narrow timing windows, inject sleeps. A\n50%-flake bug is debuggable; a 1% one is not — keep raising the rate until it is.\n\n**When it won't reproduce at all.** Stop and say so. List what you tried, then ask the user for\none of: access to an environment that reproduces it; a redacted captured artifact — HAR file,\nlog dump, core dump, timestamped screen recording; or permission to add temporary production\ninstrumentation. Wait for one of those rather than reasoning your way to a cause you cannot\nobserve.\n\n**No loop, no step 2.** If you catch yourself reading code to build a theory before that\ncommand exists, stop — jumping straight to a hypothesis is the exact failure this skill exists\nto prevent.\n\nDone when you can name **one command** you have already run at least once — showing the\ninvocation and its redacted output — that is:\n\n- [ ] **Red-capable** — it drives the real bug code path and asserts the user's exact symptom,\n      so it goes red on this bug and green once fixed. Not \"runs without erroring\". For a\n      regression, red is a measurement breaching its threshold against the baseline.\n- [ ] **Deterministic** — the same verdict every run (flaky bugs: a pinned, high reproduction\n      rate).\n- [ ] **Fast** — seconds, not minutes.\n- [ ] **Agent-runnable** — runnable unattended; a human inside it only via the human-in-the-loop\n      script.\n\nOr, when it will not reproduce at all, done when you have stopped and asked — attempts listed,\none of the three asks above made.\n\n### 2. Reproduce and minimise\n\nRun the loop and watch it go **red**. Confirm it produces the failure the **user** described\nrather than a different one nearby — wrong bug, wrong fix — that it reproduces across runs, and\nthat you have captured the exact symptom for later steps to verify a fix against.\n\nThen shrink the repro to the smallest scenario that still goes red: cut inputs, callers,\nconfig, data and steps **one at a time**, re-running the loop after each cut. Keep the\nun-minimised loop as you go — step 5 re-runs it. It pays twice: fewer moving parts left to\nsuspect in step 3, and the clean regression test in step 5.\n\nOn a flaky loop, one green run after a cut is luck rather than evidence. Re-run at the\nreproduction rate you pinned in step 1 — enough runs that a still-load-bearing element would\nhave gone red — before you accept the cut.\n\n**Performance branch.** Scale is itself part of the signal: cutting data volume, iteration\ncount or concurrency takes the measurement harness under its threshold without removing a\ncause. Pin the scale that reproduces the regression, minimise around it, and treat that pinned\nscale as load-bearing.\n\nDone when the loop has gone red on the **user's** symptom across runs with that symptom\ncaptured, the un-minimised loop is still runnable, and every remaining element is load-bearing:\nremoving any one of them turns the loop green.\n\n### 3. Hypothesise\n\nGenerate **3–5 ranked hypotheses** before testing any of them — generating one at a time\nanchors you on the first plausible idea. Each must be falsifiable, stating its prediction: \"if\nX is the cause, then changing Y makes the bug disappear \u002F changing Z makes it worse\". A\nhypothesis with no prediction is a vibe: sharpen it or drop it.\n\nShow the ranked list to the user before testing. They re-rank it instantly on domain knowledge\n(\"we just deployed a change to #3\") or name the ones they have already ruled out. Don't block\non it — proceed with your own ranking if the user is away.\n\nDone when the ranked, falsifiable list exists and has been shown.\n\n### 4. Instrument\n\nEach probe maps to a specific prediction from step 3, and you **change one variable at a\ntime**. Reach for a debugger or REPL first where the environment supports one — one breakpoint\nbeats ten logs — then targeted logs at the boundaries that distinguish the hypotheses.\n\n**Tag every debug log** with a unique prefix, e.g. `[DEBUG-a4f2]`, so cleanup is a single\nsearch. Untagged logs survive; tagged logs die.\n\n**Performance branch.** For regressions, logs are usually the wrong instrument: measure against\nstep 1's baseline, then bisect. Measure first, fix second.\n\nDone when a probe has confirmed one hypothesis, or falsified all of them and sent you back to\nstep 3 with what it taught you. Three round trips without a confirmed hypothesis is the signal\nto stop, report what you have ruled out, and clean up through step 6.\n\n### 5. Fix, test-first\n\nWrite the regression test **before** the fix, but only where a **correct seam** exists: one at\nwhich the test exercises the real bug pattern as it occurs at the call site. A seam too shallow\nto replicate the chain that triggered the bug gives false confidence.\n\n**No correct seam is itself the finding.** Report it — the codebase's shape is what keeps this\nbug from being locked down. Still fix it, and say plainly that the lockdown is missing and why.\n\nWhere a correct seam does exist, **name it as the candidate and carry it into `yun-tdd`** — the\nminimised repro shows where the bug's chain actually runs, so the seam arrives with its\nevidence rather than being re-derived. A candidate is not an agreement: `yun-tdd`'s rule\nstands, and **the user** confirms the seam there before any test is written. Unlike step 3's\nranking, this gate **blocks** — a test at an unconfirmed seam is the false confidence this step\nopened by warning about. Then turn the minimised repro into a failing test at the confirmed\nseam and drive the fix through `yun-tdd`'s **red → green** loop, which owns the seam discipline\nand what makes a test worth keeping. Finally re-run step 1's loop against the original,\nun-minimised scenario.\n\n**Performance branch.** A timing assertion at a seam is a flaky test, and `yun-tdd`'s\n**red → green** loop has no baseline or threshold to drive. Step 1's measurement harness is\nalready the regression check: fix, then run it against the recorded baseline until it goes\ngreen under the threshold, and report the before and after numbers.\n\nDone when one of:\n\n- **Seam confirmed** — the regression test is written at it, it passes, and step 1's loop is\n  green against the un-minimised scenario.\n- **No correct seam** — its absence is reported with the reason the codebase's shape blocks the\n  lockdown, and step 1's loop is green against the un-minimised scenario without one.\n- **Performance regression** — step 1's measurement harness is green under its threshold against\n  the recorded baseline, with the before and after numbers reported.\n- **Seam unconfirmed, user away** — nothing is written at it. Park: report the confirmed cause\n  and the seam candidate, and leave the fix for the confirmation. The gate holds; the run ends\n  rather than stalling behind it.\n\n### 6. Clean up\n\n- [ ] Every `[DEBUG-...]` probe is removed — search the prefix.\n- [ ] Throwaway harnesses and any diagnostic test step 1 wrote are deleted — or, when the run\n      ends without a fix, moved to scratch and their paths reported, so the run that resumes\n      inherits the loop instead of rebuilding it.\n- [ ] Every captured artifact is deleted from scratch.\n- [ ] The hypothesis that proved correct — or, when step 4 aborted, the ones it ruled out — is\n      `persist`ed through `skills\u002F_shared\u002Fmemory-convention.md` with the symptom and the loop\n      that caught it, so the next run's recall starts where this one ended.\n\nThe fix lands where an attended run lands anywhere in this harness: commit it only when the\nuser asks.\n\nDone when every box is checked.\n\n## Attribution\n\nAdapted for this workspace from **Matt Pocock's `diagnosing-bugs`**\n(github.com\u002Fmattpocock\u002Fskills, MIT). Changes: renamed to the `yun-\u003Cverb>-\u003Cnoun>` convention; the\nphases become numbered steps, each closing on a **Done when** criterion; reading `GLOSSARY.md` and\nADRs becomes a `consult` through `skills\u002F_shared\u002Fdomain-convention.md`, and the correct hypothesis\na `persist` through `skills\u002F_shared\u002Fmemory-convention.md`, with a `recall` at the head to read it\nback. The write-test-watch-it-fail-apply-fix loop is handed to `yun-tdd`, the single source of\ntruth for seams and test quality. A measurement route is added to give the performance branch a\npath through step 1, and the routes are disclosed to [`LOOPS.md`](LOOPS.md). Named tools, the\n`hitl-loop` template and the `agents\u002Fopenai.yaml` wiring are dropped, since the harness runs\nagainst repos whose stack it does not choose. Upstream's \"never log everything and grep\" is\nrestated positively per this harness's negation rule, and Spanish triggers are added to the\ndescription. The loop-first discipline, the **tight** and **red** leading words, minimisation,\nranked falsifiable hypotheses, tagged probes, the performance branch and the cleanup checklist\nare preserved.\n",{"name":5,"description":7},"main","31bfd2af63052d9bb7036b02364637ba5dc70f87","skills\u002Fyun-diagnose-bug",0,"https:\u002F\u002Fgithub.com\u002Fmatiinchauspe\u002Fyunque","https:\u002F\u002Fgithub.com\u002Fmatiinchauspe","The fools will be invencible - \r\n\r\nDevelop developing thoughts","2026-11-13 04:14:30",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add matiinchauspe\u002Fyunque --skill yun-diagnose-bug",{"tool":38,"command":39},"bunx","bunx skills add matiinchauspe\u002Fyunque --skill yun-diagnose-bug",{"tool":41,"command":42},"pnpm","pnpm dlx skills add matiinchauspe\u002Fyunque --skill yun-diagnose-bug","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmatiinchauspe\u002Fyunque\u002Fblob\u002Fmain\u002Fskills\u002Fyun-diagnose-bug\u002FSKILL.md",true,[46],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"LOOPS.md","REFERENCE","参考文档",1788,"https:\u002F\u002Fgithub.com\u002Fmatiinchauspe\u002Fyunque\u002Fblob\u002Fmain\u002Fskills\u002Fyun-diagnose-bug\u002FLOOPS.md",[53,60,67,74,81,88],{"skillNo":54,"name":55,"nameZh":6,"description":56,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":57,"categories":58,"keywords":59,"publishedAt":19},"01M7AW7HDMMJ372HG8PGM5RR0Z","yun-write-wizard","Generate an interactive bash wizard that walks a human through steps only they can perform. Use when provisioning infrastructure, setting up credentials or CI secrets, walking an unfamiliar third-party dashboard, or running a one-off migration or cutover. Don't invoke this for steps the agent can perform itself.","https:\u002F\u002Fgithub.com\u002Fmatiinchauspe\u002Fyunque\u002Fblob\u002Fmain\u002Fskills\u002Fyun-write-wizard\u002FSKILL.md",[],[],{"skillNo":61,"name":62,"nameZh":6,"description":63,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":64,"categories":65,"keywords":66,"publishedAt":19},"01M7AW7AX7MH2S7GSG6JAD4PQY","yun-write-spec","Write the spec — consolidate a large task's settled decisions into one durable document before it is sliced into tickets. Use when the decisions are settled and the user wants them written up, or says \"write the spec\" \u002F \"escribí el spec\". Declines small fixes that need no spec.","https:\u002F\u002Fgithub.com\u002Fmatiinchauspe\u002Fyunque\u002Fblob\u002Fmain\u002Fskills\u002Fyun-write-spec\u002FSKILL.md",[],[],{"skillNo":68,"name":69,"nameZh":6,"description":70,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":71,"categories":72,"keywords":73,"publishedAt":19},"01M7AW76CY32E072H8FKE8PDQ5","yun-write-pr","Write a pull request title and body that show the change — its ticket, its shape, the evidence it works, and how dangerous it is to merge.","https:\u002F\u002Fgithub.com\u002Fmatiinchauspe\u002Fyunque\u002Fblob\u002Fmain\u002Fskills\u002Fyun-write-pr\u002FSKILL.md",[],[],{"skillNo":75,"name":76,"nameZh":6,"description":77,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":78,"categories":79,"keywords":80,"publishedAt":19},"01M7AW7291YQKHYGSJKGG1EVVX","yun-write-handoff","Compact the current session into a disposable handoff baton for the next agent, pointing at durable memory instead of duplicating it.","https:\u002F\u002Fgithub.com\u002Fmatiinchauspe\u002Fyunque\u002Fblob\u002Fmain\u002Fskills\u002Fyun-write-handoff\u002FSKILL.md",[],[],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":19},"01M7AW6WXBGGYX7XMWYQFMNCVJ","yun-write-agent-docs","Use when authoring or editing a document an agent reads in this workspace — a skill, a `skills\u002F_shared\u002F` contract, `AGENTS.md` or `skills\u002FREADME.md` — deciding how it is invoked, what lives in it versus a linked file, naming it, or diagnosing why a skill misfires, bloats, or fires unreliably.","https:\u002F\u002Fgithub.com\u002Fmatiinchauspe\u002Fyunque\u002Fblob\u002Fmain\u002Fskills\u002Fyun-write-agent-docs\u002FSKILL.md",[],[],{"skillNo":89,"name":90,"nameZh":6,"description":91,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":92,"categories":93,"keywords":94,"publishedAt":19},"01M7AW6Q9K91GDDWY0YJMCYSJK","yun-tdd","Test-driven development — the red → green loop, tests written only at pre-agreed seams. Use when the user wants to build a feature or fix a bug test-first, or mentions TDD or red-green. Reached by the implement flow to drive a build at a seam.","https:\u002F\u002Fgithub.com\u002Fmatiinchauspe\u002Fyunque\u002Fblob\u002Fmain\u002Fskills\u002Fyun-tdd\u002FSKILL.md",[],[]]