[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6WX0WXH3S8FJKTRN2FCS2YC":3,"skill-same:01M6WX0WXH3S8FJKTRN2FCS2YC":78},{"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":39,"linkOnly":40,"markdownContent":41,"markdownBody":42,"frontMatter":43,"refName":51,"commitSha":52,"skillDirectory":53,"forksCount":14,"repositoryHtmlUrl":54,"repositoryHomepageUrl":55,"ownerHtmlUrl":56,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":57,"install":58,"resources":71},"01M6WX0WXH3S8FJKTRN2FCS2YC","performance-profiling",null,"Performance analysis and profiling for CPU, memory, and I\u002FO — flame graphs, benchmarks, regression detection. Use when identifying performance bottlenecks, profiling CPU\u002Fmemory\u002FI\u002FO, interpreting flame graphs, setting up benchmarking or regression detection, or optimizing a slow function or endpoint.","projectious-work\u002Fprocesskit","projectious-work","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F263889726?v=4","01M6N97TJJ118WQ35CHRV8RD0E","projectious.work","01M6N97TPWJPF0BMKYQ4PBY5W8",0,"MIT","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002Farchive\u002Fprocesskit-install-20261101\u002Fcontext\u002Ftemplates\u002Fprocesskit\u002Fv0.10.0\u002Fcontext\u002Fskills\u002Fengineering\u002Fperformance-profiling\u002FSKILL.md",[],[19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38],"performance","performance profiling","profiling","when","memory","for","v1","flame","CPU","performance analysis","graphs","benchmarks","up","use","work","regression","analysis","detection","engineering","benchmarking","2026-11-10 10:12:30",false,"---\nname: performance-profiling\ndescription: |\n  Performance analysis and profiling for CPU, memory, and I\u002FO — flame graphs, benchmarks, regression detection. Use when identifying performance bottlenecks, profiling CPU\u002Fmemory\u002FI\u002FO, interpreting flame graphs, setting up benchmarking or regression detection, or optimizing a slow function or endpoint.\nmetadata:\n  processkit:\n    apiVersion: processkit.projectious.work\u002Fv1\n    id: SKILL-performance-profiling\n    version: \"1.1.0\"\n    created: 2026-04-06T00:00:00Z\n    category: engineering\n---\n\n# Performance Profiling\n\n## Intro\n\nIntuition about performance is unreliable — always profile before\noptimizing. Work the cycle: identify the symptom, measure the\nbaseline, profile the right resource, optimize the top bottleneck,\nverify the improvement. One change at a time.\n\n## Overview\n\n### Performance analysis cycle\n\nAlways work the full cycle: **Identify -> Measure -> Profile ->\nOptimize -> Verify**.\n\n1. **Identify** the concern: slow response, high memory, excessive\n   CPU, I\u002FO stalls\n2. **Measure** the baseline with concrete numbers (latency p50\u002Fp95\u002Fp99,\n   throughput, RSS)\n3. **Profile** with the right tool for the resource type — see\n   `references\u002Fprofiling-tools.md`\n4. **Optimize** the top bottleneck only; do not scatter-shot\n5. **Verify** with the same measurement from step 2\n\nNever skip measurement. Without a number you cannot tell whether the\nchange helped or hurt.\n\n### CPU profiling\n\nIdentify what the application spends CPU time on:\n\n1. Use a **sampling** profiler (not instrumenting) to minimize\n   overhead\n2. Collect a representative workload, not a trivial test case\n3. Generate a flame graph for visual analysis\n4. Look for **wide stacks** (hot functions), **deep stacks** (excess\n   abstraction), and unexpected functions (regex compilation,\n   serialization, logging)\n\nTrack **CPU time** per operation, not wall-clock (which includes I\u002FO\nwaits).\n\n### Memory profiling\n\nIdentify allocation pressure and leaks:\n\n1. Track **peak RSS** and **allocation rate**, not just current usage\n2. For leaks, take heap snapshots at intervals and diff them\n3. Look for unbounded collections, caches without eviction, event\n   listener accumulation, and reference cycles in GC languages\n4. Distinguish a leak (unbounded growth) from high-but-stable usage,\n   which may be acceptable\n\n### I\u002FO profiling\n\nIdentify disk, network, or database bottlenecks:\n\n1. Compare wall-clock time with CPU time — a large gap is I\u002FO wait\n2. For databases: enable slow query logging, look for N+1, missing\n   indexes\n3. For network: check connection pooling, DNS, TLS handshake overhead\n4. For disk: check synchronous writes, excessive fsync, unnecessary\n   file ops\n\n### Quick checklist before profiling\n\n1. What is the **user-visible symptom**? (slow page, high bill, OOM)\n2. What **resource** is constrained? (CPU, memory, disk, network, DB)\n3. **Constant or intermittent**? (steady-state vs spike-triggered)\n4. **What changed recently?** Regressions usually correlate with\n   recent deploys\n5. What is the **acceptable target**? (p99 \u003C 200 ms, RSS \u003C 512 MB —\n   you need a goal to know when you are done)\n\n## Gotchas\n\nAgent-specific failure modes — provider-neutral pause-and-self-check items:\n\n- **Optimizing before establishing a baseline measurement.** Without a concrete before number (p50\u002Fp95\u002Fp99, RSS, throughput), there is no way to tell whether the change helped, hurt, or made no difference. Establish the baseline first — then optimize — then verify with the same measurement.\n- **Profiling under trivial or synthetic load.** A profiler run against a tight loop with no I\u002FO tells you nothing about production behavior. Collect a workload representative of real traffic — replay production logs, replicate actual access patterns, or drive realistic concurrency before reading the profiler output.\n- **Chasing secondary bottlenecks before exhausting the primary one.** After a profiling pass, fix the top bottleneck first and re-profile. Fixing the second-biggest bottleneck while the primary one still saturates the resource often produces no measurable improvement.\n- **Reporting wall-clock time when the constrained resource is CPU.** A function that takes 200ms wall-clock while waiting for a database query uses almost no CPU. Profiling the wrong resource leads to wrong conclusions. Distinguish CPU time, wall-clock, I\u002FO wait, and memory allocation — they require different tools and different fixes.\n- **Declaring success after a microbenchmark without re-running under realistic load.** Verifying a change with a tight loop under no concurrency and then shipping it is not a verification. Re-run the same realistic workload used for the baseline before declaring the goal met.\n- **Fixing a performance regression without adding a regression test.** A regression fixed once will recur unless a benchmark or load-test encodes the boundary. After every performance fix, add a test that would catch reversion to the slow path.\n- **Running high-rate sampling profilers against production.** Profilers at 100% sampling rate or instrumenting every function call introduce measurable overhead that can itself degrade the system under investigation. Use a low-overhead sampling approach in production; reserve high-rate profiling for staging.\n\n## Full reference\n\n### Profiling tools by language\n\n`references\u002Fprofiling-tools.md` has the full table. Quick picks:\n\n- **Python**: `py-spy` (sampling CPU, attach to running process,\n  flame graphs); `memray` (allocations, native + Python frames);\n  `scalene` (line-level CPU + memory + GPU); `cProfile` for built-in\n  function-level\n- **Node.js**: `clinic.js` (Doctor \u002F Bubbleprof \u002F Flame); `0x` (flame\n  graphs); Chrome DevTools for heap snapshots\n- **Rust**: `cargo flamegraph` (CPU); `criterion` (statistical\n  benchmarks); `dhat` (heap); `perf` for hardware counters\n- **Go**: `pprof` for CPU\u002Fheap\u002Fgoroutine; `trace` for scheduler and\n  GC events; `benchstat` for statistical comparison\n- **JVM**: JFR (Flight Recorder) — built-in and production-safe;\n  `async-profiler` for sampling CPU\u002Falloc\u002Flock with no safepoint\n  bias; `jmap` + MAT for heap dumps\n\n### Benchmarking and regression detection\n\n1. Isolate the code under test — benchmark the function, not the\n   setup\n2. Run enough iterations for statistical significance (mean, stddev,\n   percentiles)\n3. Control external factors: disable turbo boost, pin CPU frequency,\n   close other processes\n4. For regression detection, store baseline results in version\n   control and compare on each run\n\n```bash\n# Rust criterion\ncargo bench -- --save-baseline main\n# After changes:\ncargo bench -- --baseline main\n```\n\n### Reading flame graphs\n\n- **X-axis** is alphabetical (not time); width = proportion of\n  samples\n- **Y-axis** is stack depth, read bottom-up (caller -> callee)\n- **Wide boxes at the top** are hot leaf functions — optimize first\n- **Wide boxes at the bottom** are hot callers — consider algorithmic\n  changes\n- Look for unexpected frames: GC pauses, lock contention, allocator\n  overhead\n- Use **differential** flame graphs (red\u002Fblue) to compare before\u002Fafter\n\nCommon patterns: a single tall tower means one deep call chain\ndominates (look for unnecessary abstraction); a wide plateau at the\ntop means a leaf function consuming most CPU; many thin towers with\nthe same top function is a memoization candidate; GC\u002Fmalloc frames\nindicate allocation pressure; lock\u002Fmutex frames mean contention;\nsyscall frames (`read`, `write`, `poll`) mean the workload is I\u002FO\nbound.\n\n### Common optimization patterns\n\nAfter profiling reveals the bottleneck, apply the right fix:\n\n- **Algorithmic** — O(n²) -> O(n log n). Always check this first.\n- **Batching** — combine many small operations into fewer large ones\n- **Caching** — reuse computed results, but measure the hit rate\n- **Pooling** — reuse expensive resources (connections, threads,\n  buffers)\n- **Lazy evaluation** — defer work until actually needed\n- **Data layout** — struct-of-arrays vs array-of-structs for cache\n  friendliness\n\n### General tips\n\n- Always profile a **realistic workload**, not a microbenchmark\n- Profile in an environment **close to production** — same data\n  size, same concurrency\n- Sampling profilers (`py-spy`, `async-profiler`, `perf`) are\n  preferred over instrumenting profilers in production\n- Collect profiles for **at least 30 seconds** for statistical\n  significance\n- Compare **before and after** with the same workload to validate\n  improvements\n","# Performance Profiling\n\n## Intro\n\nIntuition about performance is unreliable — always profile before\noptimizing. Work the cycle: identify the symptom, measure the\nbaseline, profile the right resource, optimize the top bottleneck,\nverify the improvement. One change at a time.\n\n## Overview\n\n### Performance analysis cycle\n\nAlways work the full cycle: **Identify -> Measure -> Profile ->\nOptimize -> Verify**.\n\n1. **Identify** the concern: slow response, high memory, excessive\n   CPU, I\u002FO stalls\n2. **Measure** the baseline with concrete numbers (latency p50\u002Fp95\u002Fp99,\n   throughput, RSS)\n3. **Profile** with the right tool for the resource type — see\n   `references\u002Fprofiling-tools.md`\n4. **Optimize** the top bottleneck only; do not scatter-shot\n5. **Verify** with the same measurement from step 2\n\nNever skip measurement. Without a number you cannot tell whether the\nchange helped or hurt.\n\n### CPU profiling\n\nIdentify what the application spends CPU time on:\n\n1. Use a **sampling** profiler (not instrumenting) to minimize\n   overhead\n2. Collect a representative workload, not a trivial test case\n3. Generate a flame graph for visual analysis\n4. Look for **wide stacks** (hot functions), **deep stacks** (excess\n   abstraction), and unexpected functions (regex compilation,\n   serialization, logging)\n\nTrack **CPU time** per operation, not wall-clock (which includes I\u002FO\nwaits).\n\n### Memory profiling\n\nIdentify allocation pressure and leaks:\n\n1. Track **peak RSS** and **allocation rate**, not just current usage\n2. For leaks, take heap snapshots at intervals and diff them\n3. Look for unbounded collections, caches without eviction, event\n   listener accumulation, and reference cycles in GC languages\n4. Distinguish a leak (unbounded growth) from high-but-stable usage,\n   which may be acceptable\n\n### I\u002FO profiling\n\nIdentify disk, network, or database bottlenecks:\n\n1. Compare wall-clock time with CPU time — a large gap is I\u002FO wait\n2. For databases: enable slow query logging, look for N+1, missing\n   indexes\n3. For network: check connection pooling, DNS, TLS handshake overhead\n4. For disk: check synchronous writes, excessive fsync, unnecessary\n   file ops\n\n### Quick checklist before profiling\n\n1. What is the **user-visible symptom**? (slow page, high bill, OOM)\n2. What **resource** is constrained? (CPU, memory, disk, network, DB)\n3. **Constant or intermittent**? (steady-state vs spike-triggered)\n4. **What changed recently?** Regressions usually correlate with\n   recent deploys\n5. What is the **acceptable target**? (p99 \u003C 200 ms, RSS \u003C 512 MB —\n   you need a goal to know when you are done)\n\n## Gotchas\n\nAgent-specific failure modes — provider-neutral pause-and-self-check items:\n\n- **Optimizing before establishing a baseline measurement.** Without a concrete before number (p50\u002Fp95\u002Fp99, RSS, throughput), there is no way to tell whether the change helped, hurt, or made no difference. Establish the baseline first — then optimize — then verify with the same measurement.\n- **Profiling under trivial or synthetic load.** A profiler run against a tight loop with no I\u002FO tells you nothing about production behavior. Collect a workload representative of real traffic — replay production logs, replicate actual access patterns, or drive realistic concurrency before reading the profiler output.\n- **Chasing secondary bottlenecks before exhausting the primary one.** After a profiling pass, fix the top bottleneck first and re-profile. Fixing the second-biggest bottleneck while the primary one still saturates the resource often produces no measurable improvement.\n- **Reporting wall-clock time when the constrained resource is CPU.** A function that takes 200ms wall-clock while waiting for a database query uses almost no CPU. Profiling the wrong resource leads to wrong conclusions. Distinguish CPU time, wall-clock, I\u002FO wait, and memory allocation — they require different tools and different fixes.\n- **Declaring success after a microbenchmark without re-running under realistic load.** Verifying a change with a tight loop under no concurrency and then shipping it is not a verification. Re-run the same realistic workload used for the baseline before declaring the goal met.\n- **Fixing a performance regression without adding a regression test.** A regression fixed once will recur unless a benchmark or load-test encodes the boundary. After every performance fix, add a test that would catch reversion to the slow path.\n- **Running high-rate sampling profilers against production.** Profilers at 100% sampling rate or instrumenting every function call introduce measurable overhead that can itself degrade the system under investigation. Use a low-overhead sampling approach in production; reserve high-rate profiling for staging.\n\n## Full reference\n\n### Profiling tools by language\n\n`references\u002Fprofiling-tools.md` has the full table. Quick picks:\n\n- **Python**: `py-spy` (sampling CPU, attach to running process,\n  flame graphs); `memray` (allocations, native + Python frames);\n  `scalene` (line-level CPU + memory + GPU); `cProfile` for built-in\n  function-level\n- **Node.js**: `clinic.js` (Doctor \u002F Bubbleprof \u002F Flame); `0x` (flame\n  graphs); Chrome DevTools for heap snapshots\n- **Rust**: `cargo flamegraph` (CPU); `criterion` (statistical\n  benchmarks); `dhat` (heap); `perf` for hardware counters\n- **Go**: `pprof` for CPU\u002Fheap\u002Fgoroutine; `trace` for scheduler and\n  GC events; `benchstat` for statistical comparison\n- **JVM**: JFR (Flight Recorder) — built-in and production-safe;\n  `async-profiler` for sampling CPU\u002Falloc\u002Flock with no safepoint\n  bias; `jmap` + MAT for heap dumps\n\n### Benchmarking and regression detection\n\n1. Isolate the code under test — benchmark the function, not the\n   setup\n2. Run enough iterations for statistical significance (mean, stddev,\n   percentiles)\n3. Control external factors: disable turbo boost, pin CPU frequency,\n   close other processes\n4. For regression detection, store baseline results in version\n   control and compare on each run\n\n```bash\n# Rust criterion\ncargo bench -- --save-baseline main\n# After changes:\ncargo bench -- --baseline main\n```\n\n### Reading flame graphs\n\n- **X-axis** is alphabetical (not time); width = proportion of\n  samples\n- **Y-axis** is stack depth, read bottom-up (caller -> callee)\n- **Wide boxes at the top** are hot leaf functions — optimize first\n- **Wide boxes at the bottom** are hot callers — consider algorithmic\n  changes\n- Look for unexpected frames: GC pauses, lock contention, allocator\n  overhead\n- Use **differential** flame graphs (red\u002Fblue) to compare before\u002Fafter\n\nCommon patterns: a single tall tower means one deep call chain\ndominates (look for unnecessary abstraction); a wide plateau at the\ntop means a leaf function consuming most CPU; many thin towers with\nthe same top function is a memoization candidate; GC\u002Fmalloc frames\nindicate allocation pressure; lock\u002Fmutex frames mean contention;\nsyscall frames (`read`, `write`, `poll`) mean the workload is I\u002FO\nbound.\n\n### Common optimization patterns\n\nAfter profiling reveals the bottleneck, apply the right fix:\n\n- **Algorithmic** — O(n²) -> O(n log n). Always check this first.\n- **Batching** — combine many small operations into fewer large ones\n- **Caching** — reuse computed results, but measure the hit rate\n- **Pooling** — reuse expensive resources (connections, threads,\n  buffers)\n- **Lazy evaluation** — defer work until actually needed\n- **Data layout** — struct-of-arrays vs array-of-structs for cache\n  friendliness\n\n### General tips\n\n- Always profile a **realistic workload**, not a microbenchmark\n- Profile in an environment **close to production** — same data\n  size, same concurrency\n- Sampling profilers (`py-spy`, `async-profiler`, `perf`) are\n  preferred over instrumenting profilers in production\n- Collect profiles for **at least 30 seconds** for statistical\n  significance\n- Compare **before and after** with the same workload to validate\n  improvements\n",{"name":5,"metadata":44,"description":50},{"processkit":45},{"id":46,"created":47,"version":48,"category":37,"apiVersion":49},"SKILL-performance-profiling",1775433600000,"1.1.0","processkit.projectious.work\u002Fv1","Performance analysis and profiling for CPU, memory, and I\u002FO — flame graphs, benchmarks, regression detection. Use when identifying performance bottlenecks, profiling CPU\u002Fmemory\u002FI\u002FO, interpreting flame graphs, setting up benchmarking or regression detection, or optimizing a slow function or endpoint.\n","main","4876a92620ee792aac2ac9a1a806322d8a572367","archive\u002Fprocesskit-install-20261101\u002Fcontext\u002Ftemplates\u002Fprocesskit\u002Fv0.10.0\u002Fcontext\u002Fskills\u002Fengineering\u002Fperformance-profiling","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit","https:\u002F\u002Fprojectious-work.github.io\u002Fprocesskit\u002F","https:\u002F\u002Fgithub.com\u002Fprojectious-work","2026-11-11 23:24:18",{"source":8,"skillName":5,"commands":59,"prompt":69,"zipAvailable":70},[60,63,66],{"tool":61,"command":62},"npx","npx skills add projectious-work\u002Fprocesskit --skill performance-profiling",{"tool":64,"command":65},"bunx","bunx skills add projectious-work\u002Fprocesskit --skill performance-profiling",{"tool":67,"command":68},"pnpm","pnpm dlx skills add projectious-work\u002Fprocesskit --skill performance-profiling","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002Farchive\u002Fprocesskit-install-20261101\u002Fcontext\u002Ftemplates\u002Fprocesskit\u002Fv0.10.0\u002Fcontext\u002Fskills\u002Fengineering\u002Fperformance-profiling\u002FSKILL.md",true,[72],{"path":73,"resourceType":74,"resourceTypeDesc":75,"sizeBytes":76,"sourceHtmlUrl":77},"references\u002Fprofiling-tools.md","REFERENCE","参考文档",4755,"https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002Farchive\u002Fprocesskit-install-20261101\u002Fcontext\u002Ftemplates\u002Fprocesskit\u002Fv0.10.0\u002Fcontext\u002Fskills\u002Fengineering\u002Fperformance-profiling\u002Freferences\u002Fprofiling-tools.md",[79,86,93,99,125,145],{"skillNo":80,"name":81,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":82,"categories":83,"keywords":84,"publishedAt":85},"01M7665G2ED3624R3X93304265","standard-ui-design","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fstandard-ui-design\u002FSKILL.md",[],[],"2026-11-12 13:55:43",{"skillNo":87,"name":88,"nameZh":6,"description":89,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":90,"categories":91,"keywords":92,"publishedAt":85},"01M76651QTC6W0HJ4AFW2MMR3V","standard-agent-use","Company standard for working with AI agents. Use when deciding what to delegate to agents, capturing open work, keeping agent memory or knowledge notes, improving an agent workflow, or introducing agent work in a team or for a client.","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fstandard-agent-use\u002FSKILL.md",[],[],{"skillNo":94,"name":95,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":96,"categories":97,"keywords":98,"publishedAt":85},"01M76650NN4T8P241021NBR64S","standard-agent-building","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fstandard-agent-building\u002FSKILL.md",[],[],{"skillNo":100,"name":101,"nameZh":6,"description":102,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":103,"categories":104,"keywords":110,"publishedAt":39},"01M6X5MTC89B4ZGXW3JVSQCPF8","skill-management","Manage Skill package manifests as v1 ontology entities.","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002Fsrc\u002Fcontext\u002Fskills\u002Fprocesskit\u002Fskill-management\u002FSKILL.md",[105],{"categoryNo":106,"categoryName":107,"taxonomyType":108,"taxonomyTypeDesc":109},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[111,112,113,25,114,115,33,116,117,118,119,120,101,121,122,123,124],"management","skill management","ontology","create","as","transition","get","manage","update","alpha","list","package","skill package","manifests",{"skillNo":126,"name":127,"nameZh":6,"description":128,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":129,"categories":130,"keywords":132,"publishedAt":39},"01M6X5MCV74NHW5MEDNHZRF2C8","schema-management","Manage generated processkit schema contracts through deterministic MCP rebuild and inspection workflows.","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002Fsrc\u002Fcontext\u002Fskills\u002Fprocesskit\u002Fschema-management\u002FSKILL.md",[131],{"categoryNo":106,"categoryName":107,"taxonomyType":108,"taxonomyTypeDesc":109},[133,134,111,135,136,137,138,139,140,141,142,143,33,117,118,120,144],"schema","schema management","contract","rebuild","contracts","workflows","wed","generated","MCP","validation","deterministic","mode",{"skillNo":146,"name":147,"nameZh":6,"description":148,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":149,"categories":150,"keywords":151,"publishedAt":39},"01M6X5M5G6YJ96AMZGSCZJ1D13","proposition-management","Record claims, risks, beliefs, world facts, WSJF estimates, and assumptions through the Proposition interface.","https:\u002F\u002Fgithub.com\u002Fprojectious-work\u002Fprocesskit\u002Fblob\u002Fmain\u002Fsrc\u002Fcontext\u002Fskills\u002Fprocesskit\u002Fproposition-management\u002FSKILL.md",[],[152,111,153,154,114,155,156,157,33,117,158,159,119,160,120,161,162,163,164,165],"proposition management","proposition","interface","claims","record","world","creation","event","WSJF","index","updates","query","facts","log"]