[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M70WY35N42VJT16W9MA8CCGX":3,"skill-same:01M70WY35N42VJT16W9MA8CCGX":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":34,"linkOnly":35,"markdownContent":36,"markdownBody":37,"frontMatter":38,"refName":39,"commitSha":40,"skillDirectory":5,"forksCount":41,"repositoryHtmlUrl":42,"repositoryHomepageUrl":6,"ownerHtmlUrl":43,"ownerBlogUrl":6,"ownerDescription":44,"updatedAt":45,"install":46,"resources":59},"01M70WY35N42VJT16W9MA8CCGX","hill-climb",null,"Optimize a defined outcome through bounded evaluation loops. Use for hill-climb requests or repeated improvement against an executable eval, benchmark, or visual comparison; not a one-pass review.","malinskibeniamin\u002Fskills","malinskibeniamin","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F9860550?v=4","01M6HR395Z9MM67BCKMW1WKTJS","Beniamin Malinski","01M6HR39AP95WH9NHANQ91FAJ0",4,"MIT","https:\u002F\u002Fgithub.com\u002Fmalinskibeniamin\u002Fskills\u002Fblob\u002Fmain\u002Fhill-climb\u002FSKILL.md",[],[19,20,21,22,23,24,25,26,27,28,29,30,31,32,33],"hill climb","loops","optimize","improvement","requests","for","outcome","Evaluation","benchmark","eval","use","comparison","visual","review","bounded","2026-11-10 10:12:30",false,"---\nname: hill-climb\ndescription: Optimize a defined outcome through bounded evaluation loops. Use for hill-climb requests or repeated improvement against an executable eval, benchmark, or visual comparison; not a one-pass review.\n---\n\nOptimize the artifact, not its judge. Prefer executable evidence; calibrate visual judgment.\n\n## Compose with existing owners\n\nLoad only applicable [owners](..\u002Fwriting-for-agents\u002FSKILL-MECHANICS.md#loading-dependencies); share one loop and budget:\n\n- [Development lifecycle](..\u002Fdevelopment-lifecycle\u002FSKILL.md) owns implementation\u002Fdelivery; [improve](..\u002Fimprove\u002FSKILL.md) stays advisory. Report\u002Fplan\u002Freview requests authorize no implementation edits.\n- [Quantify impact](..\u002Fquantify-impact\u002FSKILL.md) owns metrics, worthwhile deltas and noise.\n- For UI, [visual review](..\u002Fvisual-review\u002FSKILL.md) owns rendered evidence\u002Fspecialist routing; [dogfood](..\u002Fdogfood\u002FSKILL.md) owns behavior\u002Frecovery checks.\n- Explicitly authorized [council](..\u002Ftstack-skills\u002Fcouncil\u002FSKILL.md) or [sol-loop](..\u002Ftstack-skills\u002Fsol-loop\u002FSKILL.md) supplies read-only rounds under [host compatibility](..\u002Fshared\u002FTSTACK.md#review-evidence-for-council-and-sol-loop).\n  Otherwise review inline, never as independent. No implicit delegation; [efficient frontier](..\u002Fefficient-frontier\u002FSKILL.md) owns routing\u002Fcapacity. Helpers share the budget.\n\n## Define the hill before editing\n\nUse the existing task contract or a compact inline note:\n\n- **Objective:** beneficiary, outcome, primary metric or rubric categories, acceptance condition and minimum worthwhile improvement.\n- **Guardrails:** protected behavior, accessibility, security, edit surface and resource limits; preserve unrelated work and the user's endpoint.\n- **Verification:** evaluator command\u002Fvisual protocol, fixture, environment, original revision and best confirmed candidate (the incumbent). Retain both references.\n- **Stop:** endpoint, finite evaluation\u002Fresource budget and plateau rule. Default: three\n  candidate attempts, accepted or not; one intervention counts once regardless of file count.\n  Baseline batches, retries and confirmation all consume the evaluation\u002Fresource budget.\n  Reserve confirmation capacity before each attempt; stop searching if it cannot fit.\n  Tighter user, reviewer or host limits win; re-baselining never resets counters.\n\nIf no credible evaluator exists, build the smallest authorized verifier or report the missing prerequisite; an easy proxy is not the requested result.\nIf the baseline already meets the outcome, confirm it and stop without speculative edits.\n\n## Calibrate the evaluator\n\nChoose the applicable branch; keep its inputs fixed throughout comparable rounds:\n\n| Branch | Baseline and proof |\n|---|---|\n| Executable correctness\u002Feval | Reproduce representative failures and retain valid controls. Freeze assertions, expected answers and supported routes; keep held-out cases for confirmation when feasible. |\n| Performance\u002Fresources | Use the [optimization experiment protocol](..\u002Fquantify-impact\u002FREFERENCE.md#optimization-experiments): identical workload\u002Fenvironment, paired measurements, noise and fidelity guardrails against original and incumbent. |\n| Visual quality | Follow [visual calibration](..\u002Fvisual-review\u002FREFERENCE.md#review-calibration). Capture original, candidate and requested reference products in comparable states; preserve design intent, behavioral and accessibility gates. |\n\nFor authorized independent visual reviews, use fresh contexts with the same artifacts, rubric, intent, constraints, check results and unresolved findings.\nWithhold desired ratings, previous scores and preferred winners; use neutral candidate labels where practical. Request evidence-backed comparisons and specific defects.\nNumeric ratings are requested-only and advisory. Inline self-review is neither blind nor independent.\n\nOnly when review policy permits staging, screen with an authorized judge before the full panel. Reserve capacity for every required final reviewer on the same candidate.\nMark not-run, unavailable or stale results explicitly, never as agreement; staging grants no provider substitution or weaker final coverage.\n\nRe-evaluate an unchanged reference to detect judge drift. If unchanged inputs receive\nmaterially different judgments, investigate or report uncertainty instead of claiming a win.\nAn evaluator repair or changed rubric, fixture, model or environment needs a fresh baseline;\nnever splice incompatible scores into one trend or weaken tests to pass. Calibrate an added\nor replacement authorized judge on the unchanged references; keep per-judge results separate.\n\n## Climb\n\n1. **Hypothesize:** choose one evidenced bottleneck\u002Fquality defect; predict a small reversible improvement without breaking a guardrail.\n2. **Change:** implement only that intervention through its owner. Preserve a recoverable incumbent; freeze scoped inputs during review.\n3. **Evaluate:** run relevant checks and the frozen evaluator on the exact candidate.\n   Reuse valid evidence; record invalid samples, failures and unavailable coverage honestly.\n4. **Confirm, then keep:** require a worthwhile gain beyond noise and intact guardrails against\n   original and incumbent. Keep it provisional until a fresh fixed batch confirms it; advance\n   the incumbent only then. Failed\u002Finconclusive confirmation rejects the trial; restore only its\n   owned edits, retaining the confirmed incumbent and others' work. Ties or missing results are not gains.\n5. **Record:** keep a compact ledger: candidate, hypothesis, evidence\u002Fdelta, guardrails,\n   keep\u002Frevert, remaining budget and unresolved findings. Use ignored\u002Ftemporary storage only\n   when the run needs it; no routine audit dumps or per-round PRs.\n\nStop at confirmed acceptance, exhausted budget, two consecutive no-gain candidates, or an external\u002Freserved blocker.\nAn owner-raised target needs a new explicit contract; preserve the prior result and budget record rather than silently escalating or creating another PR\u002Fworktree.\n\n| Exit evidence | Report |\n|---|---|\n| Current confirmation meets target and every guardrail | `met`; retain confirmed incumbent. |\n| Valid evaluations fall short at budget\u002Fplateau stop | `unmet`; retain best confirmed incumbent, not the last trial. |\n| Missing\u002Finvalid evaluator, drift or unavailable required coverage | `unverified`; retain incumbent and name the missing proof. |\n\nA higher rating never resolves a substantive defect. Apply requested per-category and reference-ranking targets individually; an overall average cannot hide a missed target.\nKeep the original scope and target; diagnose invalid evaluations rather than rerolling unchanged candidates for a passing grade.\n\nFor prompt\u002Ftool-description optimization, load the [experiment protocol](references\u002Ftool-description-optimization.md).\nEstablish correctness before compression; preserve its supported matrix and regression suite.\n\n## Confirm and hand off\n\nReuse confirmation only for unchanged inputs; re-run invalidated functional\u002Fvisual checks against the original.\nDisclose missing independent coverage; consensus or ratings cannot replace verification or human approval.\n\nReport outcome\u002Ftarget status, original-to-final evidence, kept\u002Freverted experiments,\nremaining limits and why the loop stopped. Continue through the existing delivery owner\nto the user's endpoint; hill-climb itself adds no PR, merge, deployment or monitoring grant.\n\nExamples:\n- `\u002Fhill-climb improve sidebar readability against the supplied screenshots; preserve keyboard flows; 3 candidates; local only; no delegates`\n- `\u002Fhill-climb reduce tool-description tokens while preserving the existing tool-choice eval across supported routes; 3 candidates; local only; no delegates`\n\nSource: [Theo's walkthrough](https:\u002F\u002Fx.com\u002Ftheo\u002Fstatus\u002F2118465321032106235), 06:18-08:09 (review loop), 11:39-20:44 (visual comparison, staged reviews, score drift), 26:37-30:27 (failure-derived evals, correctness before compression). Budgets and permission boundaries are harness adaptations.\n","Optimize the artifact, not its judge. Prefer executable evidence; calibrate visual judgment.\n\n## Compose with existing owners\n\nLoad only applicable [owners](..\u002Fwriting-for-agents\u002FSKILL-MECHANICS.md#loading-dependencies); share one loop and budget:\n\n- [Development lifecycle](..\u002Fdevelopment-lifecycle\u002FSKILL.md) owns implementation\u002Fdelivery; [improve](..\u002Fimprove\u002FSKILL.md) stays advisory. Report\u002Fplan\u002Freview requests authorize no implementation edits.\n- [Quantify impact](..\u002Fquantify-impact\u002FSKILL.md) owns metrics, worthwhile deltas and noise.\n- For UI, [visual review](..\u002Fvisual-review\u002FSKILL.md) owns rendered evidence\u002Fspecialist routing; [dogfood](..\u002Fdogfood\u002FSKILL.md) owns behavior\u002Frecovery checks.\n- Explicitly authorized [council](..\u002Ftstack-skills\u002Fcouncil\u002FSKILL.md) or [sol-loop](..\u002Ftstack-skills\u002Fsol-loop\u002FSKILL.md) supplies read-only rounds under [host compatibility](..\u002Fshared\u002FTSTACK.md#review-evidence-for-council-and-sol-loop).\n  Otherwise review inline, never as independent. No implicit delegation; [efficient frontier](..\u002Fefficient-frontier\u002FSKILL.md) owns routing\u002Fcapacity. Helpers share the budget.\n\n## Define the hill before editing\n\nUse the existing task contract or a compact inline note:\n\n- **Objective:** beneficiary, outcome, primary metric or rubric categories, acceptance condition and minimum worthwhile improvement.\n- **Guardrails:** protected behavior, accessibility, security, edit surface and resource limits; preserve unrelated work and the user's endpoint.\n- **Verification:** evaluator command\u002Fvisual protocol, fixture, environment, original revision and best confirmed candidate (the incumbent). Retain both references.\n- **Stop:** endpoint, finite evaluation\u002Fresource budget and plateau rule. Default: three\n  candidate attempts, accepted or not; one intervention counts once regardless of file count.\n  Baseline batches, retries and confirmation all consume the evaluation\u002Fresource budget.\n  Reserve confirmation capacity before each attempt; stop searching if it cannot fit.\n  Tighter user, reviewer or host limits win; re-baselining never resets counters.\n\nIf no credible evaluator exists, build the smallest authorized verifier or report the missing prerequisite; an easy proxy is not the requested result.\nIf the baseline already meets the outcome, confirm it and stop without speculative edits.\n\n## Calibrate the evaluator\n\nChoose the applicable branch; keep its inputs fixed throughout comparable rounds:\n\n| Branch | Baseline and proof |\n|---|---|\n| Executable correctness\u002Feval | Reproduce representative failures and retain valid controls. Freeze assertions, expected answers and supported routes; keep held-out cases for confirmation when feasible. |\n| Performance\u002Fresources | Use the [optimization experiment protocol](..\u002Fquantify-impact\u002FREFERENCE.md#optimization-experiments): identical workload\u002Fenvironment, paired measurements, noise and fidelity guardrails against original and incumbent. |\n| Visual quality | Follow [visual calibration](..\u002Fvisual-review\u002FREFERENCE.md#review-calibration). Capture original, candidate and requested reference products in comparable states; preserve design intent, behavioral and accessibility gates. |\n\nFor authorized independent visual reviews, use fresh contexts with the same artifacts, rubric, intent, constraints, check results and unresolved findings.\nWithhold desired ratings, previous scores and preferred winners; use neutral candidate labels where practical. Request evidence-backed comparisons and specific defects.\nNumeric ratings are requested-only and advisory. Inline self-review is neither blind nor independent.\n\nOnly when review policy permits staging, screen with an authorized judge before the full panel. Reserve capacity for every required final reviewer on the same candidate.\nMark not-run, unavailable or stale results explicitly, never as agreement; staging grants no provider substitution or weaker final coverage.\n\nRe-evaluate an unchanged reference to detect judge drift. If unchanged inputs receive\nmaterially different judgments, investigate or report uncertainty instead of claiming a win.\nAn evaluator repair or changed rubric, fixture, model or environment needs a fresh baseline;\nnever splice incompatible scores into one trend or weaken tests to pass. Calibrate an added\nor replacement authorized judge on the unchanged references; keep per-judge results separate.\n\n## Climb\n\n1. **Hypothesize:** choose one evidenced bottleneck\u002Fquality defect; predict a small reversible improvement without breaking a guardrail.\n2. **Change:** implement only that intervention through its owner. Preserve a recoverable incumbent; freeze scoped inputs during review.\n3. **Evaluate:** run relevant checks and the frozen evaluator on the exact candidate.\n   Reuse valid evidence; record invalid samples, failures and unavailable coverage honestly.\n4. **Confirm, then keep:** require a worthwhile gain beyond noise and intact guardrails against\n   original and incumbent. Keep it provisional until a fresh fixed batch confirms it; advance\n   the incumbent only then. Failed\u002Finconclusive confirmation rejects the trial; restore only its\n   owned edits, retaining the confirmed incumbent and others' work. Ties or missing results are not gains.\n5. **Record:** keep a compact ledger: candidate, hypothesis, evidence\u002Fdelta, guardrails,\n   keep\u002Frevert, remaining budget and unresolved findings. Use ignored\u002Ftemporary storage only\n   when the run needs it; no routine audit dumps or per-round PRs.\n\nStop at confirmed acceptance, exhausted budget, two consecutive no-gain candidates, or an external\u002Freserved blocker.\nAn owner-raised target needs a new explicit contract; preserve the prior result and budget record rather than silently escalating or creating another PR\u002Fworktree.\n\n| Exit evidence | Report |\n|---|---|\n| Current confirmation meets target and every guardrail | `met`; retain confirmed incumbent. |\n| Valid evaluations fall short at budget\u002Fplateau stop | `unmet`; retain best confirmed incumbent, not the last trial. |\n| Missing\u002Finvalid evaluator, drift or unavailable required coverage | `unverified`; retain incumbent and name the missing proof. |\n\nA higher rating never resolves a substantive defect. Apply requested per-category and reference-ranking targets individually; an overall average cannot hide a missed target.\nKeep the original scope and target; diagnose invalid evaluations rather than rerolling unchanged candidates for a passing grade.\n\nFor prompt\u002Ftool-description optimization, load the [experiment protocol](references\u002Ftool-description-optimization.md).\nEstablish correctness before compression; preserve its supported matrix and regression suite.\n\n## Confirm and hand off\n\nReuse confirmation only for unchanged inputs; re-run invalidated functional\u002Fvisual checks against the original.\nDisclose missing independent coverage; consensus or ratings cannot replace verification or human approval.\n\nReport outcome\u002Ftarget status, original-to-final evidence, kept\u002Freverted experiments,\nremaining limits and why the loop stopped. Continue through the existing delivery owner\nto the user's endpoint; hill-climb itself adds no PR, merge, deployment or monitoring grant.\n\nExamples:\n- `\u002Fhill-climb improve sidebar readability against the supplied screenshots; preserve keyboard flows; 3 candidates; local only; no delegates`\n- `\u002Fhill-climb reduce tool-description tokens while preserving the existing tool-choice eval across supported routes; 3 candidates; local only; no delegates`\n\nSource: [Theo's walkthrough](https:\u002F\u002Fx.com\u002Ftheo\u002Fstatus\u002F2118465321032106235), 06:18-08:09 (review loop), 11:39-20:44 (visual comparison, staged reviews, score drift), 26:37-30:27 (failure-derived evals, correctness before compression). Budgets and permission boundaries are harness adaptations.\n",{"name":5,"description":7},"main","33ce54f34e30fc6576782912ce0d571729558427",1,"https:\u002F\u002Fgithub.com\u002Fmalinskibeniamin\u002Fskills","https:\u002F\u002Fgithub.com\u002Fmalinskibeniamin","Software Engineer ","2026-11-14 18:47:37",{"source":8,"skillName":5,"commands":47,"prompt":57,"zipAvailable":58},[48,51,54],{"tool":49,"command":50},"npx","npx skills add malinskibeniamin\u002Fskills --skill hill-climb",{"tool":52,"command":53},"bunx","bunx skills add malinskibeniamin\u002Fskills --skill hill-climb",{"tool":55,"command":56},"pnpm","pnpm dlx skills add malinskibeniamin\u002Fskills --skill hill-climb","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmalinskibeniamin\u002Fskills\u002Fblob\u002Fmain\u002Fhill-climb\u002FSKILL.md",true,[60],{"path":61,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":64,"sourceHtmlUrl":65},"references\u002Ftool-description-optimization.md","REFERENCE","参考文档",1630,"https:\u002F\u002Fgithub.com\u002Fmalinskibeniamin\u002Fskills\u002Fblob\u002Fmain\u002Fhill-climb\u002Freferences\u002Ftool-description-optimization.md",[67,90,95,100,124,138],{"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":71,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":34},"01M70WXTA2JQ9SN3QZPHHCYB90","gestures","Build or fix web drag, swipe, sheet, reorder, and hold interactions: direct tracking, rubber banding, release velocity, momentum, snap points, and interruption. Not native React Native or Swift.","https:\u002F\u002Fgithub.com\u002Fmalinskibeniamin\u002Fskills\u002Fblob\u002Fmain\u002Fgestures\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fmalinskibeniamin\u002Fskills\u002Fblob\u002Fmain\u002Fgestures\u002FSKILL.md",[],[69,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89],"swift","velocity","react","swipe","direct","tracking","fix","sheet","release","build","native","web","drag","react native","snap",{"skillNo":91,"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":92,"categories":93,"keywords":94,"publishedAt":34},"01M70WKSB2PTJGVH7C9XC1FEED","https:\u002F\u002Fgithub.com\u002Fmalinskibeniamin\u002Fskills\u002Fblob\u002Fmain\u002Fcodex-skills\u002Fhill-climb\u002FSKILL.md",[],[19,22,28,27,26,25,24,23,21,20,29,33,32,31,30],{"skillNo":96,"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":97,"categories":98,"keywords":99,"publishedAt":34},"01M70WKGYKGN1J786GK8N3RGH1","https:\u002F\u002Fgithub.com\u002Fmalinskibeniamin\u002Fskills\u002Fblob\u002Fmain\u002Fcodex-skills\u002Fgestures\u002FSKILL.md",[],[69,82,78,81,80,79,75,77,84,85,76,86,87,88,89,83],{"skillNo":101,"name":102,"nameZh":6,"description":103,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":104,"sourceHtmlUrl":105,"categories":106,"keywords":107,"publishedAt":34},"01M6HS2326AZFPF8G9C2KW4AT0","improve-animations","Audit existing web UI motion and prioritize concrete improvements or self-contained implementation plans. Use for a motion roadmap; animate owns implementation and diff review.","https:\u002F\u002Fgithub.com\u002Fmalinskibeniamin\u002Fskills\u002Fblob\u002Fmain\u002Fimprove-animations\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fmalinskibeniamin\u002Fskills\u002Fblob\u002Fmain\u002Fimprove-animations\u002FSKILL.md",[],[108,109,110,24,111,112,86,113,114,115,116,32,117,118,119,29,120,121,122,123],"improve animations","improve","prioritize","audit","diff","self","diff review","web ui","ui motion","ui","roadmap","Existing","animate","motion","implementation","plans",{"skillNo":125,"name":126,"nameZh":6,"description":127,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":128,"sourceHtmlUrl":129,"categories":130,"keywords":131,"publishedAt":34},"01M6HS0AADDXBB0S8B42R0G82K","effect-ts","Use when setting up a repository that uses the Effect TypeScript library.","https:\u002F\u002Fgithub.com\u002Fmalinskibeniamin\u002Fskills\u002Fblob\u002Fmain\u002Feffect-ts\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fmalinskibeniamin\u002Fskills\u002Fblob\u002Fmain\u002Feffect-ts\u002FSKILL.md",[],[132,133,134,135,29,136,137],"effect ts","effect","up","library","typescript","repository",{"skillNo":139,"name":140,"nameZh":6,"description":141,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":142,"categories":143,"keywords":144,"publishedAt":34},"01M6HRDWAEFX9PKNCZSSXM3PQD","matt-chief-of-staff","Use the pinned Matt Pocock chief-of-staff skill with host compatibility rules.","https:\u002F\u002Fgithub.com\u002Fmalinskibeniamin\u002Fskills\u002Fblob\u002Fmain\u002Fcodex-skills\u002Fmatt-chief-of-staff\u002FSKILL.md",[],[145,146,147,148,149,150,151,29,152,153,154],"chief","chief of staff","matt chief of staff","chief-of-staff","compatibility","Rules","matt pocock","host","true","with"]