[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M797RAVB5TYW132PPTJ3M5Q5":3,"skill-same:01M797RAVB5TYW132PPTJ3M5Q5":44},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":26,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":29,"install":30,"resources":43},"01M797RAVB5TYW132PPTJ3M5Q5","contest-solver-engineering",null,"Diagnose and harden contest or bounded algorithmic solver implementations for compiler and toolchain mismatches, crashes, undefined behavior, overflow, memory-layout failures, TLE or MLE, buffering, serialization, randomness, data-structure invariants, and unsafe deadline handling. Use as an internal algorithmic-problem-solving recovery route after evidence implicates the implementation or target environment rather than the mathematical model, evaluator contract, or search-method design.","VectorSpaceLab\u002FAREX-Skill","VectorSpaceLab","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F180637102?v=4","01M7538C5WPSJGWF7KHR4EFZ9F","01M7538C6R2GFP53NPXTCZNBRT",351,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FFrontierCS\u002Falgorithmic-problem-solving\u002Fsub-skills\u002Fcontest-solver-engineering\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: contest-solver-engineering\ndescription: >-\n  Diagnose and harden contest or bounded algorithmic solver implementations for\n  compiler and toolchain mismatches, crashes, undefined behavior, overflow,\n  memory-layout failures, TLE or MLE, buffering, serialization, randomness,\n  data-structure invariants, and unsafe deadline handling. Use as an internal\n  algorithmic-problem-solving recovery route after evidence implicates the\n  implementation or target environment rather than the mathematical model,\n  evaluator contract, or search-method design.\n---\n\n# Contest Solver Engineering\n\nRecover implementation and environment failures after the algorithmic route is justified. Correct algorithm design can still score zero when the target compiler, protocol, serialization, numeric range, memory layout, or deadline is wrong.\n\n## Ownership and handoffs\n\nThis skill owns target-toolchain reproduction, numeric and memory safety, measured runtime, low-level I\u002FO, serialization, randomness, implementation invariants, and release hardening. Route:\n\n- a contradicted state, recurrence, reduction, proof, or asymptotic route to [model and route algorithms](..\u002Fmodel-and-route-algorithms\u002FSKILL.md);\n- evaluator parsing, legality, or score uncertainty that blocks the current decision, or official\u002Flocal disagreement to [checker and local evaluation](..\u002Fchecker-and-local-evaluation\u002FSKILL.md);\n- experiment design and champion\u002Fchallenger comparison to [validation and experiments](..\u002Fvalidation-and-experiments\u002FSKILL.md);\n- query strategy and hidden-state inference to [interactive problem solving](..\u002Finteractive-problem-solving\u002FSKILL.md);\n- scored construction, representation, moves, deltas, and optimizer behavior to the shared [heuristic search](..\u002F..\u002Freferences\u002Fheuristic-search.md).\n\n## 1. Reproduce the target environment\n\nRead the config and build command. Record:\n\n```\nlanguage standard and compiler version\noptimization\u002Fdebug flags and architecture flags\navailable headers, libraries, and runtime\ntime and memory limits per case\u002Fprocess\ninput\u002Foutput protocol and working directory\n```\n\nCompile early with the actual standard. Avoid assuming optional libraries, Boost components, compiler extensions, AVX\u002FAVX2, or native CPU flags exist. Intrinsics require matching target flags and judge hardware. Prefer portable scalar code unless the environment explicitly guarantees the feature and the gain is measured.\n\nUse warnings in development, for example:\n\n```\n-Wall -Wextra -Wshadow -Wconversion\n```\n\nTreat warnings involving return types, narrowing, signedness, uninitialized state, comparator requirements, and ambiguous names as correctness issues.\n\n## 2. Numeric safety\n\nDerive value bounds for every multiplication, sum, distance, count, and score.\n\n- Use `int64_t`\u002F`long long` when 32-bit can overflow; use `__int128` for intermediate products when supported by the target.\n- Check sentinel arithmetic. `INF + weight`, negating the minimum signed value, and evaluating a sentinel line can overflow before comparison.\n- Normalize negative modular residues and prove modular inverse existence.\n- Keep objective direction consistent when negating scores or costs.\n- Avoid exact equality on floating point unless values are constructed to be exact. Use scale-aware predicates and one boundary convention.\n- Keep scoring computations compatible with the official rounding order.\n\nSanitizers (`address`, `undefined`) are valuable on development tests, but use a release build for timing and verify sanitizer availability first.\n\n## 3. Memory layout\n\nEstimate bytes, not just Big-O:\n\n```\nvector capacity, node padding, allocator metadata, hash load factor,\nrecursion stack, duplicated versions, rollback logs, per-test clearing\n```\n\nAvoid enormous arrays as local stack objects. Allocate large pools statically or on the heap with checked capacity. Persistent structures need a proven maximum node count and overflow guard. Prefer flat contiguous arrays when traversal is hot; reserve vectors and reuse buffers inside search loops.\n\nDo not clear an `O(max_size)` table per tiny test when timestamps or touched indices suffice. Release or reuse per-case data according to total memory.\n\n## 4. Runtime and deadlines\n\nUse a monotonic clock such as `steady_clock`. Set an internal limit below the official time and include parsing, construction, final validation, and output.\n\n```\ndeadline = start + internal_budget\nperiodic check interval * worst move time bounds overshoot\n```\n\nMeasure timer-call cost before checking on every tiny transition. Conversely, do not run a large uninterruptible loop or subsolve without a deadline check or work cap. Host load and measurement granularity make near-limit results flaky.\n\nFor multiple cases under one process limit, allocate time adaptively but retain a guaranteed path to output for every remaining case. Keep the best legal state available if search stops immediately.\n\nProfile optimized binaries with representative data. Common hidden costs:\n\n- copying full state on every move;\n- full rescoring instead of local delta;\n- repeated allocation\u002Ffree and string construction;\n- maps\u002Fhashes where sorted vectors or arrays suffice;\n- RNG\u002Fdistributions in the hottest loop;\n- excessive logging or flushing;\n- pathological repair or recursion tails.\n\nIf the measured state\u002Ftransition count invalidates the selected complexity class rather than exposing an implementation cost, return to Model and Route Algorithms.\n\n## 5. Randomness\n\nUse a fast, reproducible generator appropriate for the task. During debugging, accept or log an explicit seed. Distinguish the program seed from instance and judge-noise seeds.\n\n- Avoid modulo bias when it can affect the method materially.\n- Check empty\u002Fsingleton ranges before sampling.\n- Keep deterministic tie-breaking when comparing algorithms.\n- Do not reseed repeatedly from low-resolution time.\n- A time-derived release seed may improve diversity, but preserve a way to reproduce failures and confirm the judge permits nondeterminism.\n\n## 6. Input\u002Foutput and serialization\n\nParse the exact token grammar and sizes. For large input, use suitable fast I\u002FO, but do not mix incompatible buffered I\u002FO APIs carelessly.\n\nFor batch output:\n\n- print exactly the required number\u002Forder of tokens or lines;\n- avoid extra debug text unless comments are explicitly permitted;\n- validate indices, uniqueness, counts, and bounds after conversion from the internal state;\n- ensure the output buffer\u002Fstring cannot grow without bound;\n- retain enough time for serialization and flush.\n\nFor interactive output:\n\n- read required initial data before issuing a query;\n- flush after every query\u002Faction that expects a response;\n- parse special termination\u002Ferror responses before updating state;\n- never print diagnostics to stdout; use stderr only if permitted;\n- do not wait for an initial token the protocol never sends;\n- avoid naming helpers so they collide with standard-library functions.\n\nWhen a concrete buffering, flush, deadlock, premature-EOF, or process-wiring symptom needs local reproduction, test the real process through pipes against a simulator. Stable coherent official interactions do not require simulator construction solely for this section; route evaluator implementation through Checker and Local Evaluation only when that diagnostic evidence exists.\n\nThis section owns low-level process, buffering, serialization, and deadline failures. Interactive Problem Solving owns the protocol contract, hypothesis model, and query strategy.\n\n### Nonzero `Wrong Answer` or non-perfect feedback in query-scored tasks\n\nA judge label does not always identify the failing layer. If official feedback gives a nonzero score but says `Wrong Answer` or otherwise reports a non-perfect result:\n\n- Treat the nonzero score as evidence of partial credit, not as proof that the candidate is fully legal or correct and not as proof that every non-perfect case contributed zero.\n- Do not assume `score = fully passed test cases \u002F total test cases` unless the official evaluator or scoring rules establish that formula. A single case labeled `Wrong Answer` or otherwise not fully correct may still contribute partial points.\n- If the per-case raw contribution, weights, rounding, clamps, or final aggregation is unknown or contradicted and that uncertainty blocks the current decision, recover it with [Checker and Local Evaluation](..\u002Fchecker-and-local-evaluation\u002FSKILL.md). When coherent official rules or results already establish the aggregation, use them without constructing a local evaluator. Optimize the established final official score rather than the binary count of fully passed cases.\n- If the public scoring formula contains a query-count term, invert it to estimate the hidden query count and compare that estimate with local query categories.\n- If the score has no query-count term, do not invent one; diagnose the actual scored objective instead.\n- Prioritize zero score, crash, protocol failure, malformed output, and hard-invalid candidates. Otherwise choose the next correction by its expected improvement to the established final score instead of insisting that every case become binary accepted first.\n\n## 7. Data structures and invariants\n\nAdvanced structures are high-risk code. For each operation state:\n\n```\nprecondition, mutation, lazy\u002Ftag composition, maintained aggregate,\nindex convention, complexity, inverse\u002Frollback behavior\n```\n\nProperty-test them independently. Important cases:\n\n- empty\u002Fsingle element and full range;\n- repeated equal keys and duplicate insert\u002Fdelete;\n- overlapping lazy updates in different orders;\n- split then merge identity;\n- link\u002Fcut only across valid components;\n- custom comparator is a strict weak ordering;\n- coordinate compression maps every queried value;\n- pools and queues cannot exceed capacity.\n\nPrefer a simpler verified structure if the advanced option does not decide the complexity bound.\n\n## 8. Development and release modes\n\nDevelopment build can include:\n\n- assertions and full invariant checks;\n- apply\u002Fundo and delta\u002Ffull-score comparisons;\n- deterministic seeds and verbose stderr traces;\n- sanitizers and slow reference paths;\n- operation counters and stage timers.\n\nRelease build should:\n\n- retain cheap guards needed for legality;\n- disable or sample expensive diagnostics;\n- bound all variable work;\n- use target-compatible flags only;\n- contain no filesystem\u002Fnetwork\u002Fprivate-data dependency unless authorized;\n- produce a valid fallback under early deadline.\n\nRun both modes on the same regression corpus, then rerun release in a clean process or container. Reused containers may retain writable temporary state, so a clean run is part of validation.\n\n## 9. Final audit\n\nBefore delivery, answer concretely:\n\n- Which compiler command succeeded?\n- Which official or local judge command succeeded?\n- What are representative and worst-observed runtime and memory?\n- What margin remains below the limit?\n- Which numeric products have explicit bounds?\n- Does every early-exit path produce legal output or terminate protocol safely?\n- Is the saved final artifact the champion?\n- Are unavailable headers, CPU features, accidental debug output, and private data access absent?\n\nComplete the applicable release gate in Validation and Experiments. Return the root implementation\u002Fenvironment cause, the exact build and run commands, focused regression results, observed resource margins, and remaining target-environment assumptions.\n","# Contest Solver Engineering\n\nRecover implementation and environment failures after the algorithmic route is justified. Correct algorithm design can still score zero when the target compiler, protocol, serialization, numeric range, memory layout, or deadline is wrong.\n\n## Ownership and handoffs\n\nThis skill owns target-toolchain reproduction, numeric and memory safety, measured runtime, low-level I\u002FO, serialization, randomness, implementation invariants, and release hardening. Route:\n\n- a contradicted state, recurrence, reduction, proof, or asymptotic route to [model and route algorithms](..\u002Fmodel-and-route-algorithms\u002FSKILL.md);\n- evaluator parsing, legality, or score uncertainty that blocks the current decision, or official\u002Flocal disagreement to [checker and local evaluation](..\u002Fchecker-and-local-evaluation\u002FSKILL.md);\n- experiment design and champion\u002Fchallenger comparison to [validation and experiments](..\u002Fvalidation-and-experiments\u002FSKILL.md);\n- query strategy and hidden-state inference to [interactive problem solving](..\u002Finteractive-problem-solving\u002FSKILL.md);\n- scored construction, representation, moves, deltas, and optimizer behavior to the shared [heuristic search](..\u002F..\u002Freferences\u002Fheuristic-search.md).\n\n## 1. Reproduce the target environment\n\nRead the config and build command. Record:\n\n```\nlanguage standard and compiler version\noptimization\u002Fdebug flags and architecture flags\navailable headers, libraries, and runtime\ntime and memory limits per case\u002Fprocess\ninput\u002Foutput protocol and working directory\n```\n\nCompile early with the actual standard. Avoid assuming optional libraries, Boost components, compiler extensions, AVX\u002FAVX2, or native CPU flags exist. Intrinsics require matching target flags and judge hardware. Prefer portable scalar code unless the environment explicitly guarantees the feature and the gain is measured.\n\nUse warnings in development, for example:\n\n```\n-Wall -Wextra -Wshadow -Wconversion\n```\n\nTreat warnings involving return types, narrowing, signedness, uninitialized state, comparator requirements, and ambiguous names as correctness issues.\n\n## 2. Numeric safety\n\nDerive value bounds for every multiplication, sum, distance, count, and score.\n\n- Use `int64_t`\u002F`long long` when 32-bit can overflow; use `__int128` for intermediate products when supported by the target.\n- Check sentinel arithmetic. `INF + weight`, negating the minimum signed value, and evaluating a sentinel line can overflow before comparison.\n- Normalize negative modular residues and prove modular inverse existence.\n- Keep objective direction consistent when negating scores or costs.\n- Avoid exact equality on floating point unless values are constructed to be exact. Use scale-aware predicates and one boundary convention.\n- Keep scoring computations compatible with the official rounding order.\n\nSanitizers (`address`, `undefined`) are valuable on development tests, but use a release build for timing and verify sanitizer availability first.\n\n## 3. Memory layout\n\nEstimate bytes, not just Big-O:\n\n```\nvector capacity, node padding, allocator metadata, hash load factor,\nrecursion stack, duplicated versions, rollback logs, per-test clearing\n```\n\nAvoid enormous arrays as local stack objects. Allocate large pools statically or on the heap with checked capacity. Persistent structures need a proven maximum node count and overflow guard. Prefer flat contiguous arrays when traversal is hot; reserve vectors and reuse buffers inside search loops.\n\nDo not clear an `O(max_size)` table per tiny test when timestamps or touched indices suffice. Release or reuse per-case data according to total memory.\n\n## 4. Runtime and deadlines\n\nUse a monotonic clock such as `steady_clock`. Set an internal limit below the official time and include parsing, construction, final validation, and output.\n\n```\ndeadline = start + internal_budget\nperiodic check interval * worst move time bounds overshoot\n```\n\nMeasure timer-call cost before checking on every tiny transition. Conversely, do not run a large uninterruptible loop or subsolve without a deadline check or work cap. Host load and measurement granularity make near-limit results flaky.\n\nFor multiple cases under one process limit, allocate time adaptively but retain a guaranteed path to output for every remaining case. Keep the best legal state available if search stops immediately.\n\nProfile optimized binaries with representative data. Common hidden costs:\n\n- copying full state on every move;\n- full rescoring instead of local delta;\n- repeated allocation\u002Ffree and string construction;\n- maps\u002Fhashes where sorted vectors or arrays suffice;\n- RNG\u002Fdistributions in the hottest loop;\n- excessive logging or flushing;\n- pathological repair or recursion tails.\n\nIf the measured state\u002Ftransition count invalidates the selected complexity class rather than exposing an implementation cost, return to Model and Route Algorithms.\n\n## 5. Randomness\n\nUse a fast, reproducible generator appropriate for the task. During debugging, accept or log an explicit seed. Distinguish the program seed from instance and judge-noise seeds.\n\n- Avoid modulo bias when it can affect the method materially.\n- Check empty\u002Fsingleton ranges before sampling.\n- Keep deterministic tie-breaking when comparing algorithms.\n- Do not reseed repeatedly from low-resolution time.\n- A time-derived release seed may improve diversity, but preserve a way to reproduce failures and confirm the judge permits nondeterminism.\n\n## 6. Input\u002Foutput and serialization\n\nParse the exact token grammar and sizes. For large input, use suitable fast I\u002FO, but do not mix incompatible buffered I\u002FO APIs carelessly.\n\nFor batch output:\n\n- print exactly the required number\u002Forder of tokens or lines;\n- avoid extra debug text unless comments are explicitly permitted;\n- validate indices, uniqueness, counts, and bounds after conversion from the internal state;\n- ensure the output buffer\u002Fstring cannot grow without bound;\n- retain enough time for serialization and flush.\n\nFor interactive output:\n\n- read required initial data before issuing a query;\n- flush after every query\u002Faction that expects a response;\n- parse special termination\u002Ferror responses before updating state;\n- never print diagnostics to stdout; use stderr only if permitted;\n- do not wait for an initial token the protocol never sends;\n- avoid naming helpers so they collide with standard-library functions.\n\nWhen a concrete buffering, flush, deadlock, premature-EOF, or process-wiring symptom needs local reproduction, test the real process through pipes against a simulator. Stable coherent official interactions do not require simulator construction solely for this section; route evaluator implementation through Checker and Local Evaluation only when that diagnostic evidence exists.\n\nThis section owns low-level process, buffering, serialization, and deadline failures. Interactive Problem Solving owns the protocol contract, hypothesis model, and query strategy.\n\n### Nonzero `Wrong Answer` or non-perfect feedback in query-scored tasks\n\nA judge label does not always identify the failing layer. If official feedback gives a nonzero score but says `Wrong Answer` or otherwise reports a non-perfect result:\n\n- Treat the nonzero score as evidence of partial credit, not as proof that the candidate is fully legal or correct and not as proof that every non-perfect case contributed zero.\n- Do not assume `score = fully passed test cases \u002F total test cases` unless the official evaluator or scoring rules establish that formula. A single case labeled `Wrong Answer` or otherwise not fully correct may still contribute partial points.\n- If the per-case raw contribution, weights, rounding, clamps, or final aggregation is unknown or contradicted and that uncertainty blocks the current decision, recover it with [Checker and Local Evaluation](..\u002Fchecker-and-local-evaluation\u002FSKILL.md). When coherent official rules or results already establish the aggregation, use them without constructing a local evaluator. Optimize the established final official score rather than the binary count of fully passed cases.\n- If the public scoring formula contains a query-count term, invert it to estimate the hidden query count and compare that estimate with local query categories.\n- If the score has no query-count term, do not invent one; diagnose the actual scored objective instead.\n- Prioritize zero score, crash, protocol failure, malformed output, and hard-invalid candidates. Otherwise choose the next correction by its expected improvement to the established final score instead of insisting that every case become binary accepted first.\n\n## 7. Data structures and invariants\n\nAdvanced structures are high-risk code. For each operation state:\n\n```\nprecondition, mutation, lazy\u002Ftag composition, maintained aggregate,\nindex convention, complexity, inverse\u002Frollback behavior\n```\n\nProperty-test them independently. Important cases:\n\n- empty\u002Fsingle element and full range;\n- repeated equal keys and duplicate insert\u002Fdelete;\n- overlapping lazy updates in different orders;\n- split then merge identity;\n- link\u002Fcut only across valid components;\n- custom comparator is a strict weak ordering;\n- coordinate compression maps every queried value;\n- pools and queues cannot exceed capacity.\n\nPrefer a simpler verified structure if the advanced option does not decide the complexity bound.\n\n## 8. Development and release modes\n\nDevelopment build can include:\n\n- assertions and full invariant checks;\n- apply\u002Fundo and delta\u002Ffull-score comparisons;\n- deterministic seeds and verbose stderr traces;\n- sanitizers and slow reference paths;\n- operation counters and stage timers.\n\nRelease build should:\n\n- retain cheap guards needed for legality;\n- disable or sample expensive diagnostics;\n- bound all variable work;\n- use target-compatible flags only;\n- contain no filesystem\u002Fnetwork\u002Fprivate-data dependency unless authorized;\n- produce a valid fallback under early deadline.\n\nRun both modes on the same regression corpus, then rerun release in a clean process or container. Reused containers may retain writable temporary state, so a clean run is part of validation.\n\n## 9. Final audit\n\nBefore delivery, answer concretely:\n\n- Which compiler command succeeded?\n- Which official or local judge command succeeded?\n- What are representative and worst-observed runtime and memory?\n- What margin remains below the limit?\n- Which numeric products have explicit bounds?\n- Does every early-exit path produce legal output or terminate protocol safely?\n- Is the saved final artifact the champion?\n- Are unavailable headers, CPU features, accidental debug output, and private data access absent?\n\nComplete the applicable release gate in Validation and Experiments. Return the root implementation\u002Fenvironment cause, the exact build and run commands, focused regression results, observed resource margins, and remaining target-environment assumptions.\n",{"name":5,"description":7},"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Ftask-oriented\u002FFrontierCS\u002Falgorithmic-problem-solving\u002Fsub-skills\u002Fcontest-solver-engineering",27,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab","2026-09-03 10:25:21",{"source":8,"skillName":5,"commands":31,"prompt":41,"zipAvailable":42},[32,35,38],{"tool":33,"command":34},"npx","npx skills add VectorSpaceLab\u002FAREX-Skill --skill contest-solver-engineering",{"tool":36,"command":37},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill contest-solver-engineering",{"tool":39,"command":40},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill contest-solver-engineering","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FFrontierCS\u002Falgorithmic-problem-solving\u002Fsub-skills\u002Fcontest-solver-engineering\u002FSKILL.md",true,[],[45,52,59,66,73,80],{"skillNo":46,"name":47,"nameZh":6,"description":48,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":49,"categories":50,"keywords":51,"publishedAt":18},"01M799NYSH1R3RVAHSM2D88Y84","passnet-triton-opt","Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block\u002Fgrid\u002Fwarp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-triton-opt\u002FSKILL.md",[],[],{"skillNo":53,"name":54,"nameZh":6,"description":55,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":56,"categories":57,"keywords":58,"publishedAt":18},"01M799NS5APEC0N1FZ7XZVW6BN","passnet-solve","END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-solve\u002FSKILL.md",[],[],{"skillNo":60,"name":61,"nameZh":6,"description":62,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":63,"categories":64,"keywords":65,"publishedAt":18},"01M799NNY9BKSM21V5R7QVTBJX","passnet-skill","PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under .\u002Fpass_dir\u002F, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-skill\u002FSKILL.md",[],[],{"skillNo":67,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"publishedAt":18},"01M799NHH6NP7Z1RK77KGXAN6S","passnet-pattern-fusion","Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()\u002Freplacement_args()\u002Freplacement_func() files, when a pass \"failed to match\", or when deciding how to fuse multiple kernels\u002Fops.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-pattern-fusion\u002FSKILL.md",[],[],{"skillNo":74,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":77,"categories":78,"keywords":79,"publishedAt":18},"01M799ND2J4WNC2PGAZ0A06NDF","passnet-orchestrate","Round\u002Fbatch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep\u002Frevert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-orchestrate\u002FSKILL.md",[],[],{"skillNo":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"publishedAt":18},"01M799N8SBTMY02ZDG2AZF5ZWE","passnet-feedback","Fast, deterministic feedback for PassNet work: pre-flight pattern\u002Fmatch verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-feedback\u002FSKILL.md",[],[]]