[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M79H2HD3DRHTTRVV368XJM4D":3,"skill-same:01M79H2HD3DRHTTRVV368XJM4D":87},{"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":30,"commitSha":31,"skillDirectory":32,"forksCount":14,"repositoryHtmlUrl":33,"repositoryHomepageUrl":6,"ownerHtmlUrl":34,"ownerBlogUrl":6,"ownerDescription":35,"updatedAt":36,"install":37,"resources":50},"01M79H2HD3DRHTTRVV368XJM4D","customizing-tools",null,"Decides how to acquire functionality that already exists elsewhere: accept a dependency, vendor it, cherry-pick, re-implement, take design inspiration, or learn process lessons from a postmortem. Also covers migrating a stable Python module to Rust via pyo3. Use for \"should we add this as a dependency\", \"should we vendor or fork this\", \"is this safe to depend on\", \"not invented here\", \"migrate this to Rust\", or before adopting any package, crate or library.","maraxen\u002Fcisternal","maraxen","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F67157875?v=4","01M79H1VQ5D3ZK82KNJYXD3HRH","Marielle Russo","01M79H1VQX3KYSD9DVKVWJDN7Z",0,"MIT","https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002F.praxia\u002Fskills\u002Fcustomizing-tools\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: customizing-tools\ndescription: \"Decides how to acquire functionality that already exists elsewhere: accept a dependency, vendor it, cherry-pick, re-implement, take design inspiration, or learn process lessons from a postmortem. Also covers migrating a stable Python module to Rust via pyo3. Use for \\\"should we add this as a dependency\\\", \\\"should we vendor or fork this\\\", \\\"is this safe to depend on\\\", \\\"not invented here\\\", \\\"migrate this to Rust\\\", or before adopting any package, crate or library.\"\ncovers_families: []\ndistinctive_strings:\n  - \"cherry-pick\"\n  - \"re-implement\"\n  - \"pyo3\"\n  - \"decision-matrix\"\n---\n\n# Customizing tools\n\n## When to Use\n\nUse for deciding how to acquire external functionality (accept dependency, vendor, cherry-pick, re-implement, inspiration, lessons) or before migrating a Python module to Rust via pyo3.\n\n## Scope\n\nSix ways exist to acquire functionality that already lives somewhere else: accept a dependency,\nvendor it, cherry-pick from it, re-implement it, derive inspiration from it, or derive lessons\nfrom it. Treating \"just add the dependency\" or \"let's just write our own\" as a reflex — rather\nthan a choice among six real options with real, cited tradeoffs — is exactly the failure mode\nthis skill exists to prevent. It covers two workflows: adjudicating that six-way decision, and\n(one specific, high-stakes instance of \"re-implement\") automating a Rust migration behind a pyo3\ninterface.\n\n## Workflow 1: the customization decision\n\n### Fast path — most decisions\n\nName the specific external functionality needed precisely (a package name and version, or the\nexact behavior\u002Falgorithm) — a vague need (\"something like lodash\") can't be adjudicated. Then\nread `references\u002Fdecision-matrix.md` for that option's `Recommend when` \u002F `Avoid when` bullets.\nThis document already reflects independent research (grounded, cited arguments for and against\neach option) adjudicated by three jurors reasoning through different lenses — security\u002F\nmaintenance, delivery speed, license\u002FIP. For a routine decision, matching the situation against\nthose bullets *is* the workflow; don't re-litigate a decision this document already settled.\n\n### Fresh jury — novel or high-stakes decisions\n\nRun a fresh jury instead of the fast path when the situation doesn't cleanly match any bullet in\nthe reference doc, or when the choice will be expensive to reverse (a core dependency, a\nsecurity-boundary component, a rewrite that will consume weeks). Dispatch three independent\njudgments of the *specific* situation, each through one of the three lenses the reference doc\nuses (security\u002Fmaintenance burden, delivery speed, license\u002FIP), each grounded in the situation's\nactual specifics rather than the pre-computed examples. Aggregate by looking for where the three\nlenses agree — that's a high-confidence verdict — and by reading where they disagree, since\ndisagreement usually marks a genuine tradeoff rather than noise to average away.\n\nIf a praxia workspace is available (an MCP tool resolving to `rig_run` with `flow=\"jury\"`, or the\n`write_jury_verdict` tool), prefer it for the fresh-jury step — it is praxia's existing n:1 jury\nprimitive with deterministic aggregation, purpose-built for exactly this \"multiple independent\nverdicts reduced by an aggregate\" shape. Fall back to three independent subagent dispatches\n(the `Agent` tool, or equivalent) when praxia isn't present — this skill must work standalone,\nsince cisternal is consumed by projects that don't all have praxia installed.\n\n### Record the decision\n\nOnce a decision is made — fast path or fresh jury — record it:\n\n```bash\nuv run python scripts\u002Frecord_decision.py decision \\\n    --option accept_dependency --target \"httpx for the outbound HTTP client\" \\\n    --confidence 0.8 --note \"actively maintained, MIT, small transitive closure\"\n```\n\nThis emits a real event into cisternal's own telemetry pipeline (`cisternal.telemetry`) — the\nsame substrate every cisternal-family tool already streams through — not a bespoke log file for\nthis skill alone. That is what makes these decisions queryable and auditable later: a year from\nnow, \"why did we vendor X instead of depending on it\" has an answer in the same place as every\nother cisternal-family telemetry event, not a comment someone has to remember to go find. Keep\nthe recorded rationale's citations attached to wherever the decision itself gets written down\n(a commit message, a design doc, a code comment near the integration point) — a decision like\n\"we vendored this\" is much more useful to a future reader with \"because left-pad\" attached than\nwithout it.\n\n## Workflow 2: Rust migration automation\n\nRe-implementing a stable Python module in Rust behind a pyo3 interface is itself a `reimplement`\ndecision — start by consulting the decision matrix's `reimplement` entry for the specific module\nin question before treating a migration as a foregone conclusion. Once the decision is made,\nthis workflow scaffolds and instruments the migration.\n\n### 1. Prioritize by stability, not by hotness\n\nPrefer migrating a **stable** surface: one with low recent churn, an already-narrow public API\nboundary, and existing test coverage a Rust reimplementation can be graded against. A Python\nmodule still under active design iteration is a poor migration candidate regardless of how much\nit would benefit from Rust's performance — the migration cost compounds every time the Python\nAPI it's shadowing changes underneath it. Grade parity against the existing Python test suite\nbefore cutover, the same discipline `jax-port`-style migrations already use elsewhere in this\ntool family: a graded parity test that fails on a real behavioral difference, not a rewrite that\nmerely \"looks right.\"\n\n### 2. Scaffold with pyo3 + maturin\n\nRead `references\u002Fpyo3.md` for the mixed Python\u002FRust project layout, the `#[pymodule]`\u002F\n`#[pyfunction]`\u002F`#[pyclass]` macros, GIL\u002Ffree-threading interaction (`Python::attach`\u002F\n`Python::detach`), Rust `Result` → Python exception translation, and the `abi3` stable-ABI\ndecision that determines whether one wheel covers every supported Python version or the CI\nmatrix needs one build per version. Choose the `abi3` floor *before* scaffolding — raising it\nlater means rebuilding the whole CI matrix, not editing one file.\n\n### 3. Choose the parallelism strategy\n\nFor CPU-bound hot paths, read `references\u002Frayon.md` (the default choice — the most\nbattle-tested, widest-audited option, with a well-documented `Python::allow_threads`\u002F\n`Python::detach` GIL-release pattern) and `references\u002Forx-parallel.md` (prefer this specifically\nfor longer chained iterator pipelines or early-exit-heavy searches, where its pull-based\nchunking measurably outperforms rayon's split\u002Fjoin model in its own published benchmarks — not\nas a default, as a targeted choice for that workload shape). Both integrate with pyo3 the same\nway: convert Python inputs to owned Rust data, release the GIL for the parallel region, never\nhold a `Python\u003C'_>` token across the release boundary into worker closures.\n\n### 4. Consider GPU offload and dual compilation targets\n\nFor compute-heavy kernels beyond what CPU parallelism reaches, read `references\u002Fwebgpu.md` —\n`wgpu` gives one compute-shader API surface that runs natively (Vulkan\u002FMetal\u002FDX12) and compiles\nto WASM, so the same kernel can serve a native pyo3 build and a browser\u002Fsandboxed-plugin build\nwithout a second implementation. If browser or sandboxed-plugin distribution matters at all for\nthis surface, read `references\u002Fwasm.md` before finalizing the crate structure — a crate that\nneeds to build for both `abi3` (native, pyo3) and `wasm32-unknown-unknown` has real structural\nconstraints (feature-gating, no arbitrary filesystem\u002Fthreads in the WASM sandbox) that are much\ncheaper to design in from the start than to retrofit after the pyo3-only version ships.\n\n### 5. Instrument every stage\n\nRecord each migration stage as it happens, the same way as a decision:\n\n```bash\nuv run python scripts\u002Frecord_decision.py migration-stage \\\n    --stage scaffold --target \"cisternal.export.write\" \\\n    --note \"maturin mixed layout, abi3-py313 floor\"\n```\n\nValid stages: `assess`, `scaffold`, `build`, `parity_test`, `cutover`. This is what makes the\nmigration itself a telemetered pipeline rather than a rewrite whose progress lives only in\nsomeone's memory — dogfooding cisternal's own stated purpose as the shared telemetry substrate\nfor the tool family on the tool family's own migration work.\n\n## Keeping the references current\n\nThe five Rust-topic reference docs (`pyo3.md`, `orx-parallel.md`, `rayon.md`, `wasm.md`,\n`webgpu.md`) each open with a machine-parseable version stamp:\n`\u003C!-- ref-crate-versions: name=version; checked=YYYY-MM-DD -->`. Run the staleness checker\nbefore trusting a reference doc's specifics on anything version-sensitive (an API that was\nrenamed, a feature that shipped since):\n\n```bash\nuv run python scripts\u002Fupdate_references.py --check\n```\n\nThis does **not** regenerate stale content — that requires real research (the same kind that\nproduced these docs originally: WebSearch\u002FWebFetch against each crate's actual current docs), not\nsomething a deterministic script can responsibly fake. It tells you *which* reference has\ndrifted so a fresh research pass can be scoped to just that one topic instead of guessing or\nredoing all five. `decision-matrix.md` is not version-stamped — it isn't tied to a crate release,\nonly to real incidents and case law, and stays current until a materially new incident or ruling\nchanges the picture.\n\n## Additional resources\n\n- **`references\u002Fdecision-matrix.md`** — the full six-option matrix: jury-adjudicated\n  recommend\u002Favoid guidance plus every supporting citation.\n- **`references\u002Fpyo3.md`** — pyo3 + maturin: project layout, macros, GIL\u002Ffree-threading,\n  error translation, packaging.\n- **`references\u002Frayon.md`** — CPU data-parallelism, the default choice.\n- **`references\u002Forx-parallel.md`** — CPU parallelism for chained\u002Fearly-exit-heavy workloads.\n- **`references\u002Fwasm.md`** — Rust → WebAssembly as a dual\u002Falternate compilation target.\n- **`references\u002Fwebgpu.md`** — `wgpu` as a compute-shader target alongside CPU parallelism.\n- **`scripts\u002Frecord_decision.py`** — emit a decision or migration-stage event into cisternal's\n  real telemetry pipeline.\n- **`scripts\u002Fupdate_references.py`** — detect (not fix) stale crate-version stamps.\n","# Customizing tools\n\n## When to Use\n\nUse for deciding how to acquire external functionality (accept dependency, vendor, cherry-pick, re-implement, inspiration, lessons) or before migrating a Python module to Rust via pyo3.\n\n## Scope\n\nSix ways exist to acquire functionality that already lives somewhere else: accept a dependency,\nvendor it, cherry-pick from it, re-implement it, derive inspiration from it, or derive lessons\nfrom it. Treating \"just add the dependency\" or \"let's just write our own\" as a reflex — rather\nthan a choice among six real options with real, cited tradeoffs — is exactly the failure mode\nthis skill exists to prevent. It covers two workflows: adjudicating that six-way decision, and\n(one specific, high-stakes instance of \"re-implement\") automating a Rust migration behind a pyo3\ninterface.\n\n## Workflow 1: the customization decision\n\n### Fast path — most decisions\n\nName the specific external functionality needed precisely (a package name and version, or the\nexact behavior\u002Falgorithm) — a vague need (\"something like lodash\") can't be adjudicated. Then\nread `references\u002Fdecision-matrix.md` for that option's `Recommend when` \u002F `Avoid when` bullets.\nThis document already reflects independent research (grounded, cited arguments for and against\neach option) adjudicated by three jurors reasoning through different lenses — security\u002F\nmaintenance, delivery speed, license\u002FIP. For a routine decision, matching the situation against\nthose bullets *is* the workflow; don't re-litigate a decision this document already settled.\n\n### Fresh jury — novel or high-stakes decisions\n\nRun a fresh jury instead of the fast path when the situation doesn't cleanly match any bullet in\nthe reference doc, or when the choice will be expensive to reverse (a core dependency, a\nsecurity-boundary component, a rewrite that will consume weeks). Dispatch three independent\njudgments of the *specific* situation, each through one of the three lenses the reference doc\nuses (security\u002Fmaintenance burden, delivery speed, license\u002FIP), each grounded in the situation's\nactual specifics rather than the pre-computed examples. Aggregate by looking for where the three\nlenses agree — that's a high-confidence verdict — and by reading where they disagree, since\ndisagreement usually marks a genuine tradeoff rather than noise to average away.\n\nIf a praxia workspace is available (an MCP tool resolving to `rig_run` with `flow=\"jury\"`, or the\n`write_jury_verdict` tool), prefer it for the fresh-jury step — it is praxia's existing n:1 jury\nprimitive with deterministic aggregation, purpose-built for exactly this \"multiple independent\nverdicts reduced by an aggregate\" shape. Fall back to three independent subagent dispatches\n(the `Agent` tool, or equivalent) when praxia isn't present — this skill must work standalone,\nsince cisternal is consumed by projects that don't all have praxia installed.\n\n### Record the decision\n\nOnce a decision is made — fast path or fresh jury — record it:\n\n```bash\nuv run python scripts\u002Frecord_decision.py decision \\\n    --option accept_dependency --target \"httpx for the outbound HTTP client\" \\\n    --confidence 0.8 --note \"actively maintained, MIT, small transitive closure\"\n```\n\nThis emits a real event into cisternal's own telemetry pipeline (`cisternal.telemetry`) — the\nsame substrate every cisternal-family tool already streams through — not a bespoke log file for\nthis skill alone. That is what makes these decisions queryable and auditable later: a year from\nnow, \"why did we vendor X instead of depending on it\" has an answer in the same place as every\nother cisternal-family telemetry event, not a comment someone has to remember to go find. Keep\nthe recorded rationale's citations attached to wherever the decision itself gets written down\n(a commit message, a design doc, a code comment near the integration point) — a decision like\n\"we vendored this\" is much more useful to a future reader with \"because left-pad\" attached than\nwithout it.\n\n## Workflow 2: Rust migration automation\n\nRe-implementing a stable Python module in Rust behind a pyo3 interface is itself a `reimplement`\ndecision — start by consulting the decision matrix's `reimplement` entry for the specific module\nin question before treating a migration as a foregone conclusion. Once the decision is made,\nthis workflow scaffolds and instruments the migration.\n\n### 1. Prioritize by stability, not by hotness\n\nPrefer migrating a **stable** surface: one with low recent churn, an already-narrow public API\nboundary, and existing test coverage a Rust reimplementation can be graded against. A Python\nmodule still under active design iteration is a poor migration candidate regardless of how much\nit would benefit from Rust's performance — the migration cost compounds every time the Python\nAPI it's shadowing changes underneath it. Grade parity against the existing Python test suite\nbefore cutover, the same discipline `jax-port`-style migrations already use elsewhere in this\ntool family: a graded parity test that fails on a real behavioral difference, not a rewrite that\nmerely \"looks right.\"\n\n### 2. Scaffold with pyo3 + maturin\n\nRead `references\u002Fpyo3.md` for the mixed Python\u002FRust project layout, the `#[pymodule]`\u002F\n`#[pyfunction]`\u002F`#[pyclass]` macros, GIL\u002Ffree-threading interaction (`Python::attach`\u002F\n`Python::detach`), Rust `Result` → Python exception translation, and the `abi3` stable-ABI\ndecision that determines whether one wheel covers every supported Python version or the CI\nmatrix needs one build per version. Choose the `abi3` floor *before* scaffolding — raising it\nlater means rebuilding the whole CI matrix, not editing one file.\n\n### 3. Choose the parallelism strategy\n\nFor CPU-bound hot paths, read `references\u002Frayon.md` (the default choice — the most\nbattle-tested, widest-audited option, with a well-documented `Python::allow_threads`\u002F\n`Python::detach` GIL-release pattern) and `references\u002Forx-parallel.md` (prefer this specifically\nfor longer chained iterator pipelines or early-exit-heavy searches, where its pull-based\nchunking measurably outperforms rayon's split\u002Fjoin model in its own published benchmarks — not\nas a default, as a targeted choice for that workload shape). Both integrate with pyo3 the same\nway: convert Python inputs to owned Rust data, release the GIL for the parallel region, never\nhold a `Python\u003C'_>` token across the release boundary into worker closures.\n\n### 4. Consider GPU offload and dual compilation targets\n\nFor compute-heavy kernels beyond what CPU parallelism reaches, read `references\u002Fwebgpu.md` —\n`wgpu` gives one compute-shader API surface that runs natively (Vulkan\u002FMetal\u002FDX12) and compiles\nto WASM, so the same kernel can serve a native pyo3 build and a browser\u002Fsandboxed-plugin build\nwithout a second implementation. If browser or sandboxed-plugin distribution matters at all for\nthis surface, read `references\u002Fwasm.md` before finalizing the crate structure — a crate that\nneeds to build for both `abi3` (native, pyo3) and `wasm32-unknown-unknown` has real structural\nconstraints (feature-gating, no arbitrary filesystem\u002Fthreads in the WASM sandbox) that are much\ncheaper to design in from the start than to retrofit after the pyo3-only version ships.\n\n### 5. Instrument every stage\n\nRecord each migration stage as it happens, the same way as a decision:\n\n```bash\nuv run python scripts\u002Frecord_decision.py migration-stage \\\n    --stage scaffold --target \"cisternal.export.write\" \\\n    --note \"maturin mixed layout, abi3-py313 floor\"\n```\n\nValid stages: `assess`, `scaffold`, `build`, `parity_test`, `cutover`. This is what makes the\nmigration itself a telemetered pipeline rather than a rewrite whose progress lives only in\nsomeone's memory — dogfooding cisternal's own stated purpose as the shared telemetry substrate\nfor the tool family on the tool family's own migration work.\n\n## Keeping the references current\n\nThe five Rust-topic reference docs (`pyo3.md`, `orx-parallel.md`, `rayon.md`, `wasm.md`,\n`webgpu.md`) each open with a machine-parseable version stamp:\n`\u003C!-- ref-crate-versions: name=version; checked=YYYY-MM-DD -->`. Run the staleness checker\nbefore trusting a reference doc's specifics on anything version-sensitive (an API that was\nrenamed, a feature that shipped since):\n\n```bash\nuv run python scripts\u002Fupdate_references.py --check\n```\n\nThis does **not** regenerate stale content — that requires real research (the same kind that\nproduced these docs originally: WebSearch\u002FWebFetch against each crate's actual current docs), not\nsomething a deterministic script can responsibly fake. It tells you *which* reference has\ndrifted so a fresh research pass can be scoped to just that one topic instead of guessing or\nredoing all five. `decision-matrix.md` is not version-stamped — it isn't tied to a crate release,\nonly to real incidents and case law, and stays current until a materially new incident or ruling\nchanges the picture.\n\n## Additional resources\n\n- **`references\u002Fdecision-matrix.md`** — the full six-option matrix: jury-adjudicated\n  recommend\u002Favoid guidance plus every supporting citation.\n- **`references\u002Fpyo3.md`** — pyo3 + maturin: project layout, macros, GIL\u002Ffree-threading,\n  error translation, packaging.\n- **`references\u002Frayon.md`** — CPU data-parallelism, the default choice.\n- **`references\u002Forx-parallel.md`** — CPU parallelism for chained\u002Fearly-exit-heavy workloads.\n- **`references\u002Fwasm.md`** — Rust → WebAssembly as a dual\u002Falternate compilation target.\n- **`references\u002Fwebgpu.md`** — `wgpu` as a compute-shader target alongside CPU parallelism.\n- **`scripts\u002Frecord_decision.py`** — emit a decision or migration-stage event into cisternal's\n  real telemetry pipeline.\n- **`scripts\u002Fupdate_references.py`** — detect (not fix) stale crate-version stamps.\n",{"name":5,"description":7,"covers_families":24,"distinctive_strings":25},[],[26,27,28,29],"cherry-pick","re-implement","pyo3","decision-matrix","main","027e7260c57f93382300e6154cfbe3b9e2db7af7",".praxia\u002Fskills\u002Fcustomizing-tools","https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal","https:\u002F\u002Fgithub.com\u002Fmaraxen","Graduate student at MIT researching evolution using liquid-handling robots","2026-11-14 05:11:32",{"source":8,"skillName":5,"commands":38,"prompt":48,"zipAvailable":49},[39,42,45],{"tool":40,"command":41},"npx","npx skills add maraxen\u002Fcisternal --skill customizing-tools",{"tool":43,"command":44},"bunx","bunx skills add maraxen\u002Fcisternal --skill customizing-tools",{"tool":46,"command":47},"pnpm","pnpm dlx skills add maraxen\u002Fcisternal --skill customizing-tools","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002F.praxia\u002Fskills\u002Fcustomizing-tools\u002FSKILL.md",true,[51,57,61,65,69,73,77,83],{"path":52,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Fdecision-matrix.md","REFERENCE","参考文档",28944,"https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002F.praxia\u002Fskills\u002Fcustomizing-tools\u002Freferences\u002Fdecision-matrix.md",{"path":58,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":59,"sourceHtmlUrl":60},"references\u002Forx-parallel.md",12755,"https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002F.praxia\u002Fskills\u002Fcustomizing-tools\u002Freferences\u002Forx-parallel.md",{"path":62,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":63,"sourceHtmlUrl":64},"references\u002Fpyo3.md",24371,"https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002F.praxia\u002Fskills\u002Fcustomizing-tools\u002Freferences\u002Fpyo3.md",{"path":66,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":67,"sourceHtmlUrl":68},"references\u002Frayon.md",24096,"https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002F.praxia\u002Fskills\u002Fcustomizing-tools\u002Freferences\u002Frayon.md",{"path":70,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":71,"sourceHtmlUrl":72},"references\u002Fwasm.md",19394,"https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002F.praxia\u002Fskills\u002Fcustomizing-tools\u002Freferences\u002Fwasm.md",{"path":74,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":75,"sourceHtmlUrl":76},"references\u002Fwebgpu.md",23614,"https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002F.praxia\u002Fskills\u002Fcustomizing-tools\u002Freferences\u002Fwebgpu.md",{"path":78,"resourceType":79,"resourceTypeDesc":80,"sizeBytes":81,"sourceHtmlUrl":82},"scripts\u002Frecord_decision.py","SCRIPT","脚本",4799,"https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002F.praxia\u002Fskills\u002Fcustomizing-tools\u002Fscripts\u002Frecord_decision.py",{"path":84,"resourceType":79,"resourceTypeDesc":80,"sizeBytes":85,"sourceHtmlUrl":86},"scripts\u002Fupdate_references.py",7697,"https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002F.praxia\u002Fskills\u002Fcustomizing-tools\u002Fscripts\u002Fupdate_references.py",[88,94,99,105,111,118],{"skillNo":89,"name":90,"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":91,"categories":92,"keywords":93,"publishedAt":19},"01M79H40AS0TPHWKGT8KCN2JPS","demo-skill","https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fmanifest_skill_resources\u002Fskills\u002Fdemo-skill\u002FSKILL.md",[],[],{"skillNo":95,"name":90,"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":19},"01M79H3Z6ZV62V0NZCJQ6KGSAQ","https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fmanifest_minimal\u002Fskills\u002Fdemo-skill\u002FSKILL.md",[],[],{"skillNo":100,"name":101,"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":102,"categories":103,"keywords":104,"publishedAt":19},"01M79H3Y8E6N74BSVNA01H4D5P","orchestration","https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fmanifest_dogfood_praxia\u002Fskills\u002Forchestration\u002FSKILL.md",[],[],{"skillNo":106,"name":107,"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":108,"categories":109,"keywords":110,"publishedAt":19},"01M79H3W4FGSC1CHGPB0Y9TTP9","handoff","https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002Ftests\u002Ffixtures\u002Fmanifest_dogfood_praxia\u002Fskills\u002Fhandoff\u002FSKILL.md",[],[],{"skillNo":112,"name":113,"nameZh":6,"description":114,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":115,"categories":116,"keywords":117,"publishedAt":19},"01M79H3HQP8SDNMPGQAHWV92JE","writing-skill-triggers","Use `writing-skill-triggers` when writing or shortening a skill's description or when_to_use, when a skill fails to fire or fires on the wrong task, or when `cisternal assets validate` warns about listing length.","https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002F.praxia\u002Fskills\u002Fwriting-skill-triggers\u002FSKILL.md",[],[],{"skillNo":119,"name":120,"nameZh":6,"description":121,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":122,"categories":123,"keywords":124,"publishedAt":19},"01M79H3E2J63BHKWDHDZ4YSD3J","triage-tool","Use when auditing agent MCP\u002Fskill use or pruning unused tool servers before changing project settings.","https:\u002F\u002Fgithub.com\u002Fmaraxen\u002Fcisternal\u002Fblob\u002Fmain\u002F.praxia\u002Fskills\u002Ftriage-tool\u002FSKILL.md",[],[]]