[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7FE74FSQBC8M650GSE2137X":3,"skill-same:01M7FE74FSQBC8M650GSE2137X":47},{"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":30,"ownerDescription":31,"updatedAt":32,"install":33,"resources":46},"01M7FE74FSQBC8M650GSE2137X","principle-take-small-verified-steps",null,"Advances work in small steps, each proven on the real artifact with the narrowest check that can go red; runs the full suite once, on the way into the default branch; backs every number with a measurement. Use when planning or executing multi-step work, choosing which tests to run, optimising or benchmarking, and before declaring any task done.","Med-Karim-Ben-Boubaker\u002Fkarim-skills","Med-Karim-Ben-Boubaker","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F63352044?v=4","01M7FE4PT628J1M137P2SH4M5M","Karim Ben Boubaker","01M7FE4PV4JZRQT69G3TNBR5D0",1,"MIT","https:\u002F\u002Fgithub.com\u002FMed-Karim-Ben-Boubaker\u002Fkarim-skills\u002Fblob\u002Fmain\u002Fskills\u002Fprinciple-take-small-verified-steps\u002FSKILL.md",[],[],"2026-11-16 10:05:34",false,"---\nname: principle-take-small-verified-steps\ndescription: \"Advances work in small steps, each proven on the real artifact with the narrowest check that can go red; runs the full suite once, on the way into the default branch; backs every number with a measurement. Use when planning or executing multi-step work, choosing which tests to run, optimising or benchmarking, and before declaring any task done.\"\n---\n\n# Small Verified Steps\n\nThe chance of a defect grows with the size of a change, and you know a behaviour only as far as you have checked it. So work in **baby steps**, each the smallest move visibly toward the goal and each ending in a state you can check: **orient, step, learn**, and repeat. Check each step with the narrowest test that can catch what you just changed: a full run proves nothing a targeted one didn't, and costs minutes every time.\n\n- **Slice vertically.** Each step is a **tracer bullet**: a thin path through every layer, growing from a **walking skeleton**. A working system grows from a simpler working system (**Gall's law**).\n- **Prove it on the real artifact.** Run the feature the way a person would, as the repo's run recipe says (`### Run`), writing only to its sandbox; read the actual value, open the actual screenshot or output. \"It compiles\", a passing mock and a subagent's \"done\" are claims, not proof.\n- **A check counts only if it can go red.** Break the behaviour on purpose and watch the check fail before you trust it passing. Test through the public interface so refactors keep it green (`tdd`).\n- **Build the lever.** For repetitive edits or checks, write the script, codemod or generator. It does the work the same way every time, and a reviewer can rerun it.\n- **Make the change easy, then make the easy change** (**preparatory refactoring**). Refactoring commits and behaviour commits stay separate.\n- **Stop at red.** When a step fails, fix it before building the next one on top.\n\n## Which checks to run\n\n| When | Run |\n|---|---|\n| Each red → green cycle | The one test you are working on |\n| The work is done and synced with its target branch | The tests for the code it touched (its file, module or package), in random order where the runner supports it, plus typecheck and lint on changed files, once. After a later sync or fix, rerun only the tests that change touched |\n| A merge request into the default branch is ready | The full suite and e2e, once: in the merge request's pipeline where the repo has one, locally otherwise |\n| A merge into an integration branch | Nothing more: no full suite, no e2e, no pipeline |\n| A prompt, model, model setting or tool changed and the user wants it measured | That component's evals, smallest case set first (`principle-let-models-judge`) |\n\n- A passing targeted run on unchanged code needs no second, wider run \"to make sure\".\n- Code tests never call a model, the network or a GPU: they use fakes, fixtures and recorded answers. Model behaviour is measured by evals, and evals are not tests.\n- A slow test is a defect: fix it as the FIRST section of `tdd` says. Report a slow suite with its time and its ten slowest tests, and mark the tests that need real I\u002FO so targeted runs can skip them.\n\n## Numbers\n\nOptimise only what a measurement shows matters. Work in this order: fix the bug, build the missing tool, measure, then optimise. No tuning loops on a harness that still has the bug.\n\n- **Estimate first.** A **back-of-the-envelope** figure from known costs tells you what the result should be. A measurement far from the estimate is measuring something else.\n- **Find the bottleneck.** Profile (CPU profiler, **flame graph**, query plan, the **USE method**) and name the one resource or code path that limits the result. By **Amdahl's law**, speeding up a part helps only by that part's share.\n- **Rule out impostors:** errors, cached or skipped work, a debug build, an untuned baseline, the load generator as the bottleneck.\n- **Keep the evidence with the number:** what ran, how many runs, the spread when it decides the question, and the source of every figure.\n\n## Reporting done\n\nState what was run and what was observed. Quote output as the command printed it, and the test runner's own summary line: a run of zero tests, or with skips you can't explain, proves nothing.\n\n```\nDone: CSV export skips empty rows.\nProof: test_export_skips_empty_rows failed before the fix (3 rows != 2) and passes after;\n`app export --in sample.csv` writes 2 rows. Touched tests: 14 passed in 0.8s.\n```\n\n## When it bends\n\nA **spike** exists to learn, not to ship: use the `prototype` skill and throw the code away.\n\n**Terms:** baby steps, spike, vertical slice, tracer bullet, walking skeleton, Gall's law, red-green-refactor, FIRST tests, mutation testing, test impact analysis, preparatory refactoring, atomic commits, premature optimisation, critical path, back-of-the-envelope estimation, profiling, bottleneck, Amdahl's law, percentiles, tail latency.\n","# Small Verified Steps\n\nThe chance of a defect grows with the size of a change, and you know a behaviour only as far as you have checked it. So work in **baby steps**, each the smallest move visibly toward the goal and each ending in a state you can check: **orient, step, learn**, and repeat. Check each step with the narrowest test that can catch what you just changed: a full run proves nothing a targeted one didn't, and costs minutes every time.\n\n- **Slice vertically.** Each step is a **tracer bullet**: a thin path through every layer, growing from a **walking skeleton**. A working system grows from a simpler working system (**Gall's law**).\n- **Prove it on the real artifact.** Run the feature the way a person would, as the repo's run recipe says (`### Run`), writing only to its sandbox; read the actual value, open the actual screenshot or output. \"It compiles\", a passing mock and a subagent's \"done\" are claims, not proof.\n- **A check counts only if it can go red.** Break the behaviour on purpose and watch the check fail before you trust it passing. Test through the public interface so refactors keep it green (`tdd`).\n- **Build the lever.** For repetitive edits or checks, write the script, codemod or generator. It does the work the same way every time, and a reviewer can rerun it.\n- **Make the change easy, then make the easy change** (**preparatory refactoring**). Refactoring commits and behaviour commits stay separate.\n- **Stop at red.** When a step fails, fix it before building the next one on top.\n\n## Which checks to run\n\n| When | Run |\n|---|---|\n| Each red → green cycle | The one test you are working on |\n| The work is done and synced with its target branch | The tests for the code it touched (its file, module or package), in random order where the runner supports it, plus typecheck and lint on changed files, once. After a later sync or fix, rerun only the tests that change touched |\n| A merge request into the default branch is ready | The full suite and e2e, once: in the merge request's pipeline where the repo has one, locally otherwise |\n| A merge into an integration branch | Nothing more: no full suite, no e2e, no pipeline |\n| A prompt, model, model setting or tool changed and the user wants it measured | That component's evals, smallest case set first (`principle-let-models-judge`) |\n\n- A passing targeted run on unchanged code needs no second, wider run \"to make sure\".\n- Code tests never call a model, the network or a GPU: they use fakes, fixtures and recorded answers. Model behaviour is measured by evals, and evals are not tests.\n- A slow test is a defect: fix it as the FIRST section of `tdd` says. Report a slow suite with its time and its ten slowest tests, and mark the tests that need real I\u002FO so targeted runs can skip them.\n\n## Numbers\n\nOptimise only what a measurement shows matters. Work in this order: fix the bug, build the missing tool, measure, then optimise. No tuning loops on a harness that still has the bug.\n\n- **Estimate first.** A **back-of-the-envelope** figure from known costs tells you what the result should be. A measurement far from the estimate is measuring something else.\n- **Find the bottleneck.** Profile (CPU profiler, **flame graph**, query plan, the **USE method**) and name the one resource or code path that limits the result. By **Amdahl's law**, speeding up a part helps only by that part's share.\n- **Rule out impostors:** errors, cached or skipped work, a debug build, an untuned baseline, the load generator as the bottleneck.\n- **Keep the evidence with the number:** what ran, how many runs, the spread when it decides the question, and the source of every figure.\n\n## Reporting done\n\nState what was run and what was observed. Quote output as the command printed it, and the test runner's own summary line: a run of zero tests, or with skips you can't explain, proves nothing.\n\n```\nDone: CSV export skips empty rows.\nProof: test_export_skips_empty_rows failed before the fix (3 rows != 2) and passes after;\n`app export --in sample.csv` writes 2 rows. Touched tests: 14 passed in 0.8s.\n```\n\n## When it bends\n\nA **spike** exists to learn, not to ship: use the `prototype` skill and throw the code away.\n\n**Terms:** baby steps, spike, vertical slice, tracer bullet, walking skeleton, Gall's law, red-green-refactor, FIRST tests, mutation testing, test impact analysis, preparatory refactoring, atomic commits, premature optimisation, critical path, back-of-the-envelope estimation, profiling, bottleneck, Amdahl's law, percentiles, tail latency.\n",{"name":5,"description":7},"main","03f9278801a451ca8b532ae5be7195296143861a","skills\u002Fprinciple-take-small-verified-steps",0,"https:\u002F\u002Fgithub.com\u002FMed-Karim-Ben-Boubaker\u002Fkarim-skills","https:\u002F\u002Fgithub.com\u002FMed-Karim-Ben-Boubaker","https:\u002F\u002Fkarimbenboubaker.me\u002F","AI Engineer","2026-11-10 22:54:19",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add Med-Karim-Ben-Boubaker\u002Fkarim-skills --skill principle-take-small-verified-steps",{"tool":39,"command":40},"bunx","bunx skills add Med-Karim-Ben-Boubaker\u002Fkarim-skills --skill principle-take-small-verified-steps",{"tool":42,"command":43},"pnpm","pnpm dlx skills add Med-Karim-Ben-Boubaker\u002Fkarim-skills --skill principle-take-small-verified-steps","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FMed-Karim-Ben-Boubaker\u002Fkarim-skills\u002Fblob\u002Fmain\u002Fskills\u002Fprinciple-take-small-verified-steps\u002FSKILL.md",true,[],[48,55,62,69,76,83],{"skillNo":49,"name":50,"nameZh":6,"description":51,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":52,"categories":53,"keywords":54,"publishedAt":19},"01M7FE8Y56P42CZS85V5NWGQ0B","write-walkthrough","Guide you through running a demo yourself that shows a fix or a new feature working: a few numbered steps in the chat, each a command to copy or a click to make, with what you should see.","https:\u002F\u002Fgithub.com\u002FMed-Karim-Ben-Boubaker\u002Fkarim-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwrite-walkthrough\u002FSKILL.md",[],[],{"skillNo":56,"name":57,"nameZh":6,"description":58,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":59,"categories":60,"keywords":61,"publishedAt":19},"01M7FE8SY1WVS77ZM15VSX9H08","write-spec","Turn the current conversation into a spec and publish it to the project issue tracker: no interview, just synthesis of what you've already discussed.","https:\u002F\u002Fgithub.com\u002FMed-Karim-Ben-Boubaker\u002Fkarim-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwrite-spec\u002FSKILL.md",[],[],{"skillNo":63,"name":64,"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},"01M7FE8Q2ZBJ61N6CEVFCVQPHR","write-questionnaire","Turn a decision you can't fully answer into a questionnaire for someone else to fill in.","https:\u002F\u002Fgithub.com\u002FMed-Karim-Ben-Boubaker\u002Fkarim-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwrite-questionnaire\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},"01M7FE8JMMVYP23YHD3KQ43SH4","write-pr","Writes a PR or MR from a fixed template: the tickets it closes, a summary sketch, before and after evidence that proves each acceptance criterion (screenshots for visible changes), and merge danger (one-way or two-way door, blast radius). Use when opening, writing or updating a pull request or merge request.","https:\u002F\u002Fgithub.com\u002FMed-Karim-Ben-Boubaker\u002Fkarim-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwrite-pr\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},"01M7FE8EEG42Z95804TK2KT89R","write-issue","Files and moves issues on the repo's tracker (GitLab, GitHub or local files): one kind (bug, feature, chore) and one stage each, a title, the kind's template, and what each later stage needs. Use when filing, relabelling, moving or closing an issue, when work should be recorded instead of done now, or to split big work into child tickets.","https:\u002F\u002Fgithub.com\u002FMed-Karim-Ben-Boubaker\u002Fkarim-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwrite-issue\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},"01M7FE8AJW55TTN9R3KFHBC9T5","write-for-agents","How to write documents an agent reads: context pointers, information hierarchy, completion criteria, leading words and pruning, plus skill frontmatter and invocation mechanics. Use when creating or editing a skill, AGENTS.md, CLAUDE.md, or any doc an agent reaches through a pointer.","https:\u002F\u002Fgithub.com\u002FMed-Karim-Ben-Boubaker\u002Fkarim-skills\u002Fblob\u002Fmain\u002Fskills\u002Fwrite-for-agents\u002FSKILL.md",[],[]]