[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M789ASPN7YA4AMS2EBA2SSGX":3,"skill-same:01M789ASPN7YA4AMS2EBA2SSGX":66},{"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":32,"commitSha":33,"skillDirectory":34,"forksCount":35,"repositoryHtmlUrl":36,"repositoryHomepageUrl":37,"ownerHtmlUrl":38,"ownerBlogUrl":39,"ownerDescription":40,"updatedAt":41,"install":42,"resources":55},"01M789ASPN7YA4AMS2EBA2SSGX","systematic-debugging",null,"Find the root cause of a defect before changing anything. Load when something crashes, hangs, returns the wrong value, fails intermittently, or broke after a change, and the cause is not yet known. Drives reproduce, localize by bisection, explain the mechanism, then fix with a failing-first regression test. Replaces guess-and-check editing.","sebastienrousseau\u002Fagtmls","sebastienrousseau","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F1394998?v=4","01M78952QGX0RG3V4QVGESS3N7","Sebastien Rousseau","01M78952R7PPYXEW5VZKQRB0VW",1,"MIT","https:\u002F\u002Fgithub.com\u002Fsebastienrousseau\u002Fagtmls\u002Fblob\u002Fmain\u002Fskills\u002Fsystematic-debugging\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: systematic-debugging\ndescription: \"Find the root cause of a defect before changing anything. Load when something crashes, hangs, returns the wrong value, fails intermittently, or broke after a change, and the cause is not yet known. Drives reproduce, localize by bisection, explain the mechanism, then fix with a failing-first regression test. Replaces guess-and-check editing.\"\nlicense: MIT\ncompatibility: \"Requires a reproduction command. Tested with Claude Code, Codex, and Aider skill layouts\"\nallowed-tools: \"Read Glob Grep Write Edit Bash\"\nmetadata:\n  agtmls-owner: \"Sebastien Rousseau\"\n  agtmls-maturity: \"hardened\"\n  agtmls-risk-level: \"low\"\n  agtmls-network-access: \"none\"\n  agtmls-writes-files: \"true\"\n  agtmls-executes-commands: \"true\"\n  agtmls-handles-secrets: \"false\"\n  agtmls-requires-human-review: \"true\"\n---\n\n# Systematic debugging\n\n**Trigger.** Something is broken and you do not yet know why: a crash, a\nwrong result, a hang, a flake, a test that fails only in CI, a regression\nafter an upgrade. Also load it when you catch yourself about to change code\nin the hope that it helps.\n\n> **Reference material** (localization techniques per symptom class,\n> intermittent-failure playbook, and the instrumentation recipes): see\n> `reference.md` in this directory.\n\n## The rule\n\n> **No edit before an explanation.** You must be able to say *this input\n> reaches this line, which does this wrong thing, which produces this\n> symptom* — before you touch the code.\n\nGuess-and-check feels faster and is not. Each speculative edit changes the\nsystem you are trying to understand, and a \"fix\" that removes a symptom\nwithout an explanation has usually moved the bug somewhere less visible.\n\n## Four phases\n\nDo not skip forward. Each phase has an exit condition; if you cannot meet it,\nyou are still in that phase.\n\n### Phase 1 — Reproduce\n\n**Exit condition: one command that fails, reliably, on demand.**\n\n1. Get the exact failure: full error, stack trace, exit code, wrong output.\n   Not a paraphrase — the literal text.\n2. Establish the environment: branch, commit, dependency versions, config,\n   platform. Bugs that \"only happen for them\" are usually environment deltas.\n3. Reduce the trigger to the smallest input and shortest command you can.\n4. Run it three times. If it fails 3\u002F3, continue. If it fails 1\u002F3, you have\n   an intermittent bug — see `reference.md` §intermittent before proceeding.\n\nIf you cannot reproduce it, **say so and stop**. Debugging a failure you\ncannot observe is writing fiction. Ask for the input, the logs, the version.\n\n### Phase 2 — Localize\n\n**Exit condition: the smallest region of code where correct input goes in\nand incorrect output comes out.**\n\nBisection is the workhorse — over *anything* ordered:\n\n| Axis | Method |\n| --- | --- |\n| Time | `git bisect` between a known-good and known-bad commit |\n| Call stack | Assert or log the invariant at the top, middle, and bottom |\n| Input | Halve the input until the failure disappears |\n| Config | Disable half the flags\u002Fplugins; repeat on the failing half |\n| Data | Halve the dataset; find the minimal failing record |\n\nThe discipline is the same every time: **form a binary question, answer it by\nobservation, halve the space.** Ten halvings cover a thousand candidates.\n\nTwo rules that keep this honest:\n\n- **Verify the boundary.** When you think you have found the transition\n  point, confirm both sides: correct just before, incorrect just after.\n- **Instrument, don't infer.** Print the actual value. A variable you\n  *believe* holds `3` is the reason the bug survived this long.\n\n### Phase 3 — Explain\n\n**Exit condition: a written sentence that connects cause to symptom, which\npredicts something you have not yet tested.**\n\nWrite it out:\n\n> *When `parse()` receives a key containing `:`, the split on line 88 takes\n> the first segment only, so the map is built with a truncated key, so the\n> lookup on line 140 misses and returns the default.*\n\nThen **test the prediction**. A correct explanation predicts more than the\none failure you have: it says which *other* inputs fail, and which\nsuperficially similar ones do not. Check one of each. If the prediction is\nwrong, your explanation is wrong — return to Phase 2.\n\nThis is the step people skip, and skipping it is how you get a fix that\nworks on the reported input and nothing else.\n\n### Phase 4 — Fix and pin\n\n**Exit condition: a regression test observed failing before the fix and\npassing after.**\n\n1. **Write the test first**, from the explanation, and watch it fail with the\n   real symptom — not `ImportError`, not a missing fixture.\n2. Make the smallest change the explanation calls for. If the explanation\n   does not call for it, it is not part of the fix.\n3. Watch the test pass.\n4. Run the full suite. A fix that breaks two other things is not a fix.\n5. Check the explanation for **siblings**: if `split(':')` was wrong here, is\n   the same pattern wrong in the three other places it appears?\n\n## Anti-patterns\n\n- **Shotgun editing.** Changing several things at once, then testing. If it\n  works you have learned nothing; if it does not, you have a bigger space.\n- **Fixing the symptom.** Wrapping the crash in try\u002Fexcept, clamping the bad\n  value, adding a retry. Legitimate only when the explanation says the\n  condition is genuinely expected — and then it is not a bug fix.\n- **Blaming the platform.** The compiler, the runtime, the standard library.\n  Occasionally right; assume it last, and only with a minimal reproduction.\n- **Trusting the comment.** Comments describe intent at writing time. Read\n  the code.\n- **Stopping at \"it stopped happening.\"** Especially for intermittent bugs.\n  Absence of a symptom over ten runs is weak evidence about a race.\n- **Rewriting to make the bug go away.** Now you have an unexplained bug and\n  a large diff.\n\n## When not to use\n\n- The project has a symptom-indexed triage playbook for its own subsystems —\n  use that first; it will localize faster than a generic bisection.\n- You already know the cause and are just making the change.\n- The task is adding new behaviour, not restoring intended behaviour — that\n  is `test-driven-development`.\n","# Systematic debugging\n\n**Trigger.** Something is broken and you do not yet know why: a crash, a\nwrong result, a hang, a flake, a test that fails only in CI, a regression\nafter an upgrade. Also load it when you catch yourself about to change code\nin the hope that it helps.\n\n> **Reference material** (localization techniques per symptom class,\n> intermittent-failure playbook, and the instrumentation recipes): see\n> `reference.md` in this directory.\n\n## The rule\n\n> **No edit before an explanation.** You must be able to say *this input\n> reaches this line, which does this wrong thing, which produces this\n> symptom* — before you touch the code.\n\nGuess-and-check feels faster and is not. Each speculative edit changes the\nsystem you are trying to understand, and a \"fix\" that removes a symptom\nwithout an explanation has usually moved the bug somewhere less visible.\n\n## Four phases\n\nDo not skip forward. Each phase has an exit condition; if you cannot meet it,\nyou are still in that phase.\n\n### Phase 1 — Reproduce\n\n**Exit condition: one command that fails, reliably, on demand.**\n\n1. Get the exact failure: full error, stack trace, exit code, wrong output.\n   Not a paraphrase — the literal text.\n2. Establish the environment: branch, commit, dependency versions, config,\n   platform. Bugs that \"only happen for them\" are usually environment deltas.\n3. Reduce the trigger to the smallest input and shortest command you can.\n4. Run it three times. If it fails 3\u002F3, continue. If it fails 1\u002F3, you have\n   an intermittent bug — see `reference.md` §intermittent before proceeding.\n\nIf you cannot reproduce it, **say so and stop**. Debugging a failure you\ncannot observe is writing fiction. Ask for the input, the logs, the version.\n\n### Phase 2 — Localize\n\n**Exit condition: the smallest region of code where correct input goes in\nand incorrect output comes out.**\n\nBisection is the workhorse — over *anything* ordered:\n\n| Axis | Method |\n| --- | --- |\n| Time | `git bisect` between a known-good and known-bad commit |\n| Call stack | Assert or log the invariant at the top, middle, and bottom |\n| Input | Halve the input until the failure disappears |\n| Config | Disable half the flags\u002Fplugins; repeat on the failing half |\n| Data | Halve the dataset; find the minimal failing record |\n\nThe discipline is the same every time: **form a binary question, answer it by\nobservation, halve the space.** Ten halvings cover a thousand candidates.\n\nTwo rules that keep this honest:\n\n- **Verify the boundary.** When you think you have found the transition\n  point, confirm both sides: correct just before, incorrect just after.\n- **Instrument, don't infer.** Print the actual value. A variable you\n  *believe* holds `3` is the reason the bug survived this long.\n\n### Phase 3 — Explain\n\n**Exit condition: a written sentence that connects cause to symptom, which\npredicts something you have not yet tested.**\n\nWrite it out:\n\n> *When `parse()` receives a key containing `:`, the split on line 88 takes\n> the first segment only, so the map is built with a truncated key, so the\n> lookup on line 140 misses and returns the default.*\n\nThen **test the prediction**. A correct explanation predicts more than the\none failure you have: it says which *other* inputs fail, and which\nsuperficially similar ones do not. Check one of each. If the prediction is\nwrong, your explanation is wrong — return to Phase 2.\n\nThis is the step people skip, and skipping it is how you get a fix that\nworks on the reported input and nothing else.\n\n### Phase 4 — Fix and pin\n\n**Exit condition: a regression test observed failing before the fix and\npassing after.**\n\n1. **Write the test first**, from the explanation, and watch it fail with the\n   real symptom — not `ImportError`, not a missing fixture.\n2. Make the smallest change the explanation calls for. If the explanation\n   does not call for it, it is not part of the fix.\n3. Watch the test pass.\n4. Run the full suite. A fix that breaks two other things is not a fix.\n5. Check the explanation for **siblings**: if `split(':')` was wrong here, is\n   the same pattern wrong in the three other places it appears?\n\n## Anti-patterns\n\n- **Shotgun editing.** Changing several things at once, then testing. If it\n  works you have learned nothing; if it does not, you have a bigger space.\n- **Fixing the symptom.** Wrapping the crash in try\u002Fexcept, clamping the bad\n  value, adding a retry. Legitimate only when the explanation says the\n  condition is genuinely expected — and then it is not a bug fix.\n- **Blaming the platform.** The compiler, the runtime, the standard library.\n  Occasionally right; assume it last, and only with a minimal reproduction.\n- **Trusting the comment.** Comments describe intent at writing time. Read\n  the code.\n- **Stopping at \"it stopped happening.\"** Especially for intermittent bugs.\n  Absence of a symptom over ten runs is weak evidence about a race.\n- **Rewriting to make the bug go away.** Now you have an unexplained bug and\n  a large diff.\n\n## When not to use\n\n- The project has a symptom-indexed triage playbook for its own subsystems —\n  use that first; it will localize faster than a generic bisection.\n- You already know the cause and are just making the change.\n- The task is adding new behaviour, not restoring intended behaviour — that\n  is `test-driven-development`.\n",{"name":5,"license":15,"metadata":24,"description":7,"allowed-tools":30,"compatibility":31},{"agtmls-owner":12,"agtmls-maturity":25,"agtmls-risk-level":26,"agtmls-writes-files":27,"agtmls-network-access":28,"agtmls-handles-secrets":29,"agtmls-executes-commands":27,"agtmls-requires-human-review":27},"hardened","low","true","none","false","Read Glob Grep Write Edit Bash","Requires a reproduction command. Tested with Claude Code, Codex, and Aider skill layouts","main","ecf7dfb9e0505c9584e5838603d8cd9cb0e64efc","skills\u002Fsystematic-debugging",0,"https:\u002F\u002Fgithub.com\u002Fsebastienrousseau\u002Fagtmls","https:\u002F\u002Fpypi.org\u002Fproject\u002Fagtmls\u002F","https:\u002F\u002Fgithub.com\u002Fsebastienrousseau","https:\u002F\u002Fsebastienrousseau.com","Payments leader, 19 years in banking. I build open-source ISO 20022, post-quantum cryptography and applied AI for finance.","2026-11-09 15:12:45",{"source":8,"skillName":5,"commands":43,"prompt":53,"zipAvailable":54},[44,47,50],{"tool":45,"command":46},"npx","npx skills add sebastienrousseau\u002Fagtmls --skill systematic-debugging",{"tool":48,"command":49},"bunx","bunx skills add sebastienrousseau\u002Fagtmls --skill systematic-debugging",{"tool":51,"command":52},"pnpm","pnpm dlx skills add sebastienrousseau\u002Fagtmls --skill systematic-debugging","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fsebastienrousseau\u002Fagtmls\u002Fblob\u002Fmain\u002Fskills\u002Fsystematic-debugging\u002FSKILL.md",true,[56,62],{"path":57,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":60,"sourceHtmlUrl":61},"metadata.json","REFERENCE","参考文档",406,"https:\u002F\u002Fgithub.com\u002Fsebastienrousseau\u002Fagtmls\u002Fblob\u002Fmain\u002Fskills\u002Fsystematic-debugging\u002Fmetadata.json",{"path":63,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":64,"sourceHtmlUrl":65},"reference.md",6020,"https:\u002F\u002Fgithub.com\u002Fsebastienrousseau\u002Fagtmls\u002Fblob\u002Fmain\u002Fskills\u002Fsystematic-debugging\u002Freference.md",[67,74,81,88,95,102],{"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},"01M789BV3SY53CRXRKQXT13DSB","example-skill","Use when the user needs this reusable workflow. Replace this with trigger-rich wording that says when to load the skill, what task it handles, and when not to use it.","https:\u002F\u002Fgithub.com\u002Fsebastienrousseau\u002Fagtmls\u002Fblob\u002Fmain\u002Ftemplates\u002Fskill\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},"01M789BPMXXTXQPSMDQ06DSV1P","writing-plans","Turn a vague or multi-step request into an ordered list of small steps, each with its own done-condition and verification. Load before starting work that touches several files, has unclear scope, or spans more than one session, and when a request needs decomposing before any code is written. Produces a plan a reviewer can check against, not a to-do list.","https:\u002F\u002Fgithub.com\u002Fsebastienrousseau\u002Fagtmls\u002Fblob\u002Fmain\u002Fskills\u002Fwriting-plans\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},"01M789BK5BBSDBQENG56CPAEDF","web-research-and-source-triage","Use when gathering current web evidence, comparing sources, triaging credibility, reading GitHub repositories, checking docs, or deciding whether a claim needs live verification.","https:\u002F\u002Fgithub.com\u002Fsebastienrousseau\u002Fagtmls\u002Fblob\u002Fmain\u002Fskills\u002Fweb-research-and-source-triage\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},"01M789BF76AAKX8MGWA48YFHMX","vetting-a-skill-before-install","Use before installing, importing or enabling a third-party skill, plugin, marketplace or skills repository in Claude Code, Codex, Antigravity, Cursor or any agent. Pins the source to a commit, audits every skill and the agent configuration it ships, reads what the audit could not cover, and ends in install, mirror or refuse. Not for auditing this registry's own skills.","https:\u002F\u002Fgithub.com\u002Fsebastienrousseau\u002Fagtmls\u002Fblob\u002Fmain\u002Fskills\u002Fvetting-a-skill-before-install\u002FSKILL.md",[],[],{"skillNo":96,"name":97,"nameZh":6,"description":98,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":99,"categories":100,"keywords":101,"publishedAt":19},"01M789B5SQ8G7RCAJBYZ0CR0R4","verification-before-completion","The gate between doing work and claiming it is done. Load before saying finished, fixed, working, or ready, before marking a task complete, and before reporting an outcome to a human. Converts each claim into the observation that would justify it, and names the rationalizations that let unverified work ship. Not the project's own evidence bar or test taxonomy.","https:\u002F\u002Fgithub.com\u002Fsebastienrousseau\u002Fagtmls\u002Fblob\u002Fmain\u002Fskills\u002Fverification-before-completion\u002FSKILL.md",[],[],{"skillNo":103,"name":104,"nameZh":6,"description":105,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":106,"categories":107,"keywords":108,"publishedAt":19},"01M789B1JXCYBRY5P3MWYQBKJ0","using-agtmls","Meta-router for the AgtMLS skill catalog. Read first when unsure which AgtMLS skill applies, when several look relevant, when asking 'is there a skill for porting or noyalib work', or when onboarding to how the hub's skills and shared references fit together.","https:\u002F\u002Fgithub.com\u002Fsebastienrousseau\u002Fagtmls\u002Fblob\u002Fmain\u002Fskills\u002Fusing-agtmls\u002FSKILL.md",[],[]]