[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6DNK9VFNJYKB6GMA15WJ3HZ":3,"skill-same:01M6DNK9VFNJYKB6GMA15WJ3HZ":75},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":38,"linkOnly":39,"markdownContent":40,"markdownBody":41,"frontMatter":42,"refName":55,"commitSha":56,"skillDirectory":57,"forksCount":58,"repositoryHtmlUrl":59,"repositoryHomepageUrl":6,"ownerHtmlUrl":60,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":61,"install":62,"resources":74},"01M6DNK9VFNJYKB6GMA15WJ3HZ","threat-model-first",null,"Use at the START of a change that moves money, changes balances or custody, touches access control or signing, deserializes untrusted input, adds an external or on-chain integration, or wires an LLM tool. Design-time threat modeling that turns abuse cases into the first failing tests; it precedes the post-hoc security-review pass.","0xgleb\u002Fdotconfig","0xgleb","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F39841057?v=4","01M6DNDD83BRMM1SH4R4C6K8GY","01M6DNDDCPH33FEFB6JN6Y2GVD",3,"MIT","https:\u002F\u002Fgithub.com\u002F0xgleb\u002Fdotconfig\u002Fblob\u002Fmaster\u002Fai\u002Fskills\u002Fthreat-model-first\u002FSKILL.md",[],[18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37],"threat model","threat","first","threat model first","model","modeling","edit","write","read","audit","change","review","cargo audit","Bash","access","forge","design","security-review","time","chain","2026-11-10 10:12:30",false,"---\nname: threat-model-first\nuser-invocable: true\nallowed-tools:\n  - Read\n  - Grep\n  - Glob\n  - Write\n  - Edit\n  - \"Bash(rg *)\"\n  - \"Bash(cargo test *)\"\n  - \"Bash(forge test *)\"\n  - \"Bash(cargo audit *)\"\n  - \"Bash(git diff --cached *)\"\ndescription: Use at the START of a change that moves money, changes balances or custody, touches access control or signing, deserializes untrusted input, adds an external or on-chain integration, or wires an LLM tool. Design-time threat modeling that turns abuse cases into the first failing tests; it precedes the post-hoc security-review pass.\n---\n\nThreat-model the change before you write it: map trust boundaries, name the\nassets, run STRIDE per boundary, and encode every abuse case as a failing test\nbefore any fix exists. This is design-time work — it runs at the start, not as\nthe security-review pass that audits a finished diff.\n\nWhen this skill activates, do the steps in order. Do not skip to implementation\nbecause the change \"looks small\" — a one-line `as u64` cast at a money boundary\nis exactly the class of bug this catches.\n\n## 1. Map the trust boundaries in THIS change\n\nList every point where data or control crosses from something you do not fully\ncontrol into your typed internal code. For this diff, that means:\n\n- Deserialized payloads (`serde` from JSON \u002F bincode \u002F a queue message \u002F a\n  webhook body).\n- Calldata and event-log decoding; `abi.decode`, generated bindings\n  reinterpreted by hand.\n- External-contract return values (an ERC-20 `transfer` return bool, a router\n  quote, a callback).\n- Oracle \u002F price feeds and anything they gate.\n- Config and environment values produced by something other than this code.\n- LLM and tool output — model completions, tool-call arguments, retrieved\n  documents.\n\nWrite the boundaries down explicitly. Everything inside a boundary is \"already\ntyped and validated\"; everything crossing one is suspect until proven.\n\n## 2. Name the assets\n\nState what an attacker would want and what an honest bug could corrupt:\n\n- Balances and custody (who can withdraw, who is credited).\n- Money movement (transfers, mints, burns, settlements, fee skims).\n- Signing keys and signature-authorized actions.\n- Admin roles and privileged setters.\n- Invariant-protected state (total supply == sum of balances, collateral ratios,\n  accounting identities).\n\nEach asset is a thing a later test must defend. If the change touches none of\nthese, this skill probably should not have activated — stop and confirm the\ntrigger.\n\n## 3. STRIDE per boundary (recast to this domain)\n\nFor each boundary from step 1, walk the six and write down the concrete attack\nand its mitigation in this codebase:\n\n| Threat                 | Question in this domain             | Mitigation lever                                                                                            |\n| ---------------------- | ----------------------------------- | ----------------------------------------------------------------------------------------------------------- |\n| Spoofing               | Can a caller forge identity?        | `msg.sender` \u002F role checks; ECDSA signature recovery and signer allowlist; nonce per signer                 |\n| Tampering              | Can a value be altered or overflow? | Type invariants (invalid states unrepresentable); checked \u002F saturating arithmetic; no silent `as` narrowing |\n| Repudiation            | Can an action be denied later?      | Emit events for every state change; structured, queryable logs                                              |\n| Information disclosure | Can a secret leak?                  | No keys \u002F seeds \u002F raw amounts in logs or revert strings; least data in errors                               |\n| Denial of service      | Can it be wedged or gas-bombed?     | Bound inputs, loops, and unbounded growth; pull-over-push payouts; timeouts on external calls               |\n| Elevation of privilege | Can a user gain rights?             | Authorization checks at the entry point; least privilege; no defaulting to admin                            |\n\n## 4. Write abuse cases as the FIRST failing tests\n\nBefore any fix exists, turn each credible attack into a test that compiles and\nfails. This is the type-driven TDD discipline applied to security: the failing\ntest is the spec for the fix.\n\n- Solidity (`forge test`): reentrancy via a malicious callback receiver,\n  unauthorized caller (`vm.prank` a non-owner), overflow \u002F underflow, rounding\n  direction, signature replay (reuse a consumed signature \u002F wrong chainid),\n  double-spend, oracle manipulation (move the feed within a block).\n- Rust (`cargo test`, proptest): malformed \u002F adversarial deserialized input,\n  unchecked arithmetic on attacker-influenced amounts, boundary widths (`u256`\n  truncated to `u64`), property invariants that must hold for all inputs.\n\nA reproduction of a vulnerability being fixed must fail for the right reason\nbefore the fix. Existing passing malformed-input and security tests for current\nretained contracts remain valuable. Apply GLOBAL_AGENTS.md's obsolete-capability\nrule: removing a capability does not require replacement tests or an inventory\nof retired names.\n\n## 5. Solidity canon\n\nApply and verify against each boundary:\n\n- Checks-effects-interactions ordering; reentrancy guard where external calls\n  follow state reads.\n- Round in the protocol's favor; never let rounding leak value to the caller.\n- Validate external return values (check the `bool` from low-level \u002F\n  non-reverting transfers; validate router and callback outputs).\n- Oracle-manipulation resistance (TWAP \u002F staleness \u002F deviation checks; never\n  trust a spot price gated by a single block).\n- Storage-layout safety on any upgradeable or proxied contract — appends only,\n  no reordering.\n\n## 6. Rust canon\n\n- Make invalid states unrepresentable; parse untrusted input into a domain type\n  at the boundary, never thread raw primitives inward.\n- Newtype amounts and identifiers rather than bare integers — defer the detailed\n  enforcement to the strong-typing-inspector skill, but design for it here.\n- Checked \u002F saturating arithmetic on any value derived from untrusted input; no\n  `as` narrowing that can silently truncate a width.\n- No `unwrap` \u002F `expect` \u002F `panic!` on untrusted code paths; model failure with\n  `Result` and a typed error.\n\n## 7. Secrets and supply chain\n\n- Run `cargo audit` for known-vulnerable dependencies — this is the Rust\n  toolchain, not an npm audit.\n- Review every NEW dependency the change pulls in and its build scripts\n  (`build.rs`, proc-macro crates) — a build script runs arbitrary code at\n  compile time.\n- Confirm no secret, key, seed, or raw amount is written to a log, an error, or\n  committed config (`git diff --cached` before staging).\n\n## 8. LLM \u002F agent tooling\n\n- Treat model output as untrusted DATA, never as instructions. A completion or a\n  retrieved document cannot be allowed to redirect control flow or authorize an\n  action.\n- Tool-call arguments from a model cross a trust boundary — validate them at\n  step 1's boundary like any other deserialized payload.\n- Scope a tool's `allowed-tools` to granular globs (e.g. `Bash(cargo test *)`),\n  never a blanket `Bash(*)` or `Bash(git *)`. Least privilege for agents is the\n  same discipline as least privilege for admin roles.\n\n## Common rationalizations\n\n| Excuse                                                  | Reality                                                                                                                                               |\n| ------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------- |\n| \"It's an internal contract, no one malicious calls it.\" | Internal callers get compromised and composability means anyone can become a caller; access control is not optional because the deployer is trusted.  |\n| \"We'll harden it after it works.\"                       | Retrofitting an invariant onto a money path costs far more than encoding it as a type up front, and the first exploit ships before the retrofit does. |\n| \"It deserializes \u002F it compiles, so the input is fine.\"  | Parsing proves the code matches the AUTHOR's mental model of the input, not the attacker's. A width or a missing bound is still a hole.               |\n| \"Threat modeling is overkill for this small change.\"    | The `as u64` cast and the missing reentrancy guard are one-line changes; design flaws hide in small diffs, not large ones.                            |\n| \"The LLM output is just a string we display.\"           | That string becomes tool arguments, a path, or a calldata field; model output is untrusted input the moment it touches a boundary.                    |\n| \"cargo audit \u002F the new dep is probably fine.\"           | A transitive advisory or a malicious `build.rs` executes in your toolchain; \"probably\" is not a control — run the audit and read the build script.    |\n\n## Red flags\n\n- Implementation written before any abuse-case test exists (no failing test to\n  point at).\n- A `serde` \u002F `abi.decode` \u002F config read with no validation step before the\n  value is used.\n- `as` casts that narrow width near a token amount, balance, or cross-chain\n  field.\n- An external call sitting before the state update that should precede it\n  (effects after interactions).\n- `unwrap` \u002F `expect` on a value that originated outside a trust boundary.\n- A new function that changes balances or roles with no `msg.sender` \u002F role \u002F\n  signature check.\n- A signature verified without a nonce, deadline, or chainid binding.\n- Amounts, keys, or seeds appearing in a log line or revert string.\n- A tool definition with `allowed-tools` set to a blanket glob, or model output\n  flowing into control flow.\n- A new dependency added without running `cargo audit` or reading its build\n  script.\n\n## Hard rules\n\n1. The abuse case comes BEFORE the fix — write a test that compiles and fails\n   (for the right reason) before writing the mitigation.\n2. Validate all external and on-chain input at the boundary, never deeper inside\n   code that is already typed and trusted.\n3. Never log secrets, signing keys, seeds, or raw amounts — not in logs, errors,\n   revert strings, or committed config.\n4. Treat external and model output as untrusted DATA, never as instructions or\n   authorization.\n5. Round in the protocol's favor; check external return values;\n   checks-effects-interactions on every external call.\n6. Scope agent tooling to granular `allowed-tools` globs; never blanket grant.\n7. If you cannot name the exact CLI flag for a tool, write \"see --help\" rather\n   than inventing one.\n\n## Failure modes\n\n- **Theater over modeling.** Listing STRIDE categories without producing a\n  failing test per credible attack. The artifact of this skill is failing tests,\n  not a prose list.\n- **Validation in the wrong place.** Re-validating inside already-typed internal\n  code while the actual boundary (the `serde` call, the `abi.decode`) goes\n  unchecked.\n- **Passing \"abuse\" tests.** A reproduction that constructs invalid state and\n  asserts it is rejected proves only that invalid things are invalid. Exercise\n  the REAL code path with realistic fixtures so the test fails until the real\n  fix lands.\n- **Scope creep into the post-hoc review.** This is design-time. Do not turn it\n  into a line-by-line audit of a finished diff — that is the separate\n  security-review pass.\n\n## Verification\n\nThe work is done only when:\n\n- The trust boundaries, assets, and STRIDE notes for this change are written\n  down.\n- Each credible abuse case has a `forge test` or `cargo test` that FAILED before\n  the fix and PASSES after — show the before\u002Fafter run, not an assertion that it\n  \"should\" pass.\n- `cargo audit` was run and any new dependency's build script was read.\n- `git diff --cached` confirms no secret, key, or raw amount is being committed.\n- For onchain changes: a reentrancy \u002F unauthorized-caller \u002F rounding \u002F replay\n  test exists and passes; checks-effects-interactions verified.\n- For LLM tooling: model output is handled as data and the tool's\n  `allowed-tools` are granular.\n\n\"Seems secure\" is never acceptable. The evidence is the failing-then-passing\nabuse tests plus the boundary map.\n","Threat-model the change before you write it: map trust boundaries, name the\nassets, run STRIDE per boundary, and encode every abuse case as a failing test\nbefore any fix exists. This is design-time work — it runs at the start, not as\nthe security-review pass that audits a finished diff.\n\nWhen this skill activates, do the steps in order. Do not skip to implementation\nbecause the change \"looks small\" — a one-line `as u64` cast at a money boundary\nis exactly the class of bug this catches.\n\n## 1. Map the trust boundaries in THIS change\n\nList every point where data or control crosses from something you do not fully\ncontrol into your typed internal code. For this diff, that means:\n\n- Deserialized payloads (`serde` from JSON \u002F bincode \u002F a queue message \u002F a\n  webhook body).\n- Calldata and event-log decoding; `abi.decode`, generated bindings\n  reinterpreted by hand.\n- External-contract return values (an ERC-20 `transfer` return bool, a router\n  quote, a callback).\n- Oracle \u002F price feeds and anything they gate.\n- Config and environment values produced by something other than this code.\n- LLM and tool output — model completions, tool-call arguments, retrieved\n  documents.\n\nWrite the boundaries down explicitly. Everything inside a boundary is \"already\ntyped and validated\"; everything crossing one is suspect until proven.\n\n## 2. Name the assets\n\nState what an attacker would want and what an honest bug could corrupt:\n\n- Balances and custody (who can withdraw, who is credited).\n- Money movement (transfers, mints, burns, settlements, fee skims).\n- Signing keys and signature-authorized actions.\n- Admin roles and privileged setters.\n- Invariant-protected state (total supply == sum of balances, collateral ratios,\n  accounting identities).\n\nEach asset is a thing a later test must defend. If the change touches none of\nthese, this skill probably should not have activated — stop and confirm the\ntrigger.\n\n## 3. STRIDE per boundary (recast to this domain)\n\nFor each boundary from step 1, walk the six and write down the concrete attack\nand its mitigation in this codebase:\n\n| Threat                 | Question in this domain             | Mitigation lever                                                                                            |\n| ---------------------- | ----------------------------------- | ----------------------------------------------------------------------------------------------------------- |\n| Spoofing               | Can a caller forge identity?        | `msg.sender` \u002F role checks; ECDSA signature recovery and signer allowlist; nonce per signer                 |\n| Tampering              | Can a value be altered or overflow? | Type invariants (invalid states unrepresentable); checked \u002F saturating arithmetic; no silent `as` narrowing |\n| Repudiation            | Can an action be denied later?      | Emit events for every state change; structured, queryable logs                                              |\n| Information disclosure | Can a secret leak?                  | No keys \u002F seeds \u002F raw amounts in logs or revert strings; least data in errors                               |\n| Denial of service      | Can it be wedged or gas-bombed?     | Bound inputs, loops, and unbounded growth; pull-over-push payouts; timeouts on external calls               |\n| Elevation of privilege | Can a user gain rights?             | Authorization checks at the entry point; least privilege; no defaulting to admin                            |\n\n## 4. Write abuse cases as the FIRST failing tests\n\nBefore any fix exists, turn each credible attack into a test that compiles and\nfails. This is the type-driven TDD discipline applied to security: the failing\ntest is the spec for the fix.\n\n- Solidity (`forge test`): reentrancy via a malicious callback receiver,\n  unauthorized caller (`vm.prank` a non-owner), overflow \u002F underflow, rounding\n  direction, signature replay (reuse a consumed signature \u002F wrong chainid),\n  double-spend, oracle manipulation (move the feed within a block).\n- Rust (`cargo test`, proptest): malformed \u002F adversarial deserialized input,\n  unchecked arithmetic on attacker-influenced amounts, boundary widths (`u256`\n  truncated to `u64`), property invariants that must hold for all inputs.\n\nA reproduction of a vulnerability being fixed must fail for the right reason\nbefore the fix. Existing passing malformed-input and security tests for current\nretained contracts remain valuable. Apply GLOBAL_AGENTS.md's obsolete-capability\nrule: removing a capability does not require replacement tests or an inventory\nof retired names.\n\n## 5. Solidity canon\n\nApply and verify against each boundary:\n\n- Checks-effects-interactions ordering; reentrancy guard where external calls\n  follow state reads.\n- Round in the protocol's favor; never let rounding leak value to the caller.\n- Validate external return values (check the `bool` from low-level \u002F\n  non-reverting transfers; validate router and callback outputs).\n- Oracle-manipulation resistance (TWAP \u002F staleness \u002F deviation checks; never\n  trust a spot price gated by a single block).\n- Storage-layout safety on any upgradeable or proxied contract — appends only,\n  no reordering.\n\n## 6. Rust canon\n\n- Make invalid states unrepresentable; parse untrusted input into a domain type\n  at the boundary, never thread raw primitives inward.\n- Newtype amounts and identifiers rather than bare integers — defer the detailed\n  enforcement to the strong-typing-inspector skill, but design for it here.\n- Checked \u002F saturating arithmetic on any value derived from untrusted input; no\n  `as` narrowing that can silently truncate a width.\n- No `unwrap` \u002F `expect` \u002F `panic!` on untrusted code paths; model failure with\n  `Result` and a typed error.\n\n## 7. Secrets and supply chain\n\n- Run `cargo audit` for known-vulnerable dependencies — this is the Rust\n  toolchain, not an npm audit.\n- Review every NEW dependency the change pulls in and its build scripts\n  (`build.rs`, proc-macro crates) — a build script runs arbitrary code at\n  compile time.\n- Confirm no secret, key, seed, or raw amount is written to a log, an error, or\n  committed config (`git diff --cached` before staging).\n\n## 8. LLM \u002F agent tooling\n\n- Treat model output as untrusted DATA, never as instructions. A completion or a\n  retrieved document cannot be allowed to redirect control flow or authorize an\n  action.\n- Tool-call arguments from a model cross a trust boundary — validate them at\n  step 1's boundary like any other deserialized payload.\n- Scope a tool's `allowed-tools` to granular globs (e.g. `Bash(cargo test *)`),\n  never a blanket `Bash(*)` or `Bash(git *)`. Least privilege for agents is the\n  same discipline as least privilege for admin roles.\n\n## Common rationalizations\n\n| Excuse                                                  | Reality                                                                                                                                               |\n| ------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------- |\n| \"It's an internal contract, no one malicious calls it.\" | Internal callers get compromised and composability means anyone can become a caller; access control is not optional because the deployer is trusted.  |\n| \"We'll harden it after it works.\"                       | Retrofitting an invariant onto a money path costs far more than encoding it as a type up front, and the first exploit ships before the retrofit does. |\n| \"It deserializes \u002F it compiles, so the input is fine.\"  | Parsing proves the code matches the AUTHOR's mental model of the input, not the attacker's. A width or a missing bound is still a hole.               |\n| \"Threat modeling is overkill for this small change.\"    | The `as u64` cast and the missing reentrancy guard are one-line changes; design flaws hide in small diffs, not large ones.                            |\n| \"The LLM output is just a string we display.\"           | That string becomes tool arguments, a path, or a calldata field; model output is untrusted input the moment it touches a boundary.                    |\n| \"cargo audit \u002F the new dep is probably fine.\"           | A transitive advisory or a malicious `build.rs` executes in your toolchain; \"probably\" is not a control — run the audit and read the build script.    |\n\n## Red flags\n\n- Implementation written before any abuse-case test exists (no failing test to\n  point at).\n- A `serde` \u002F `abi.decode` \u002F config read with no validation step before the\n  value is used.\n- `as` casts that narrow width near a token amount, balance, or cross-chain\n  field.\n- An external call sitting before the state update that should precede it\n  (effects after interactions).\n- `unwrap` \u002F `expect` on a value that originated outside a trust boundary.\n- A new function that changes balances or roles with no `msg.sender` \u002F role \u002F\n  signature check.\n- A signature verified without a nonce, deadline, or chainid binding.\n- Amounts, keys, or seeds appearing in a log line or revert string.\n- A tool definition with `allowed-tools` set to a blanket glob, or model output\n  flowing into control flow.\n- A new dependency added without running `cargo audit` or reading its build\n  script.\n\n## Hard rules\n\n1. The abuse case comes BEFORE the fix — write a test that compiles and fails\n   (for the right reason) before writing the mitigation.\n2. Validate all external and on-chain input at the boundary, never deeper inside\n   code that is already typed and trusted.\n3. Never log secrets, signing keys, seeds, or raw amounts — not in logs, errors,\n   revert strings, or committed config.\n4. Treat external and model output as untrusted DATA, never as instructions or\n   authorization.\n5. Round in the protocol's favor; check external return values;\n   checks-effects-interactions on every external call.\n6. Scope agent tooling to granular `allowed-tools` globs; never blanket grant.\n7. If you cannot name the exact CLI flag for a tool, write \"see --help\" rather\n   than inventing one.\n\n## Failure modes\n\n- **Theater over modeling.** Listing STRIDE categories without producing a\n  failing test per credible attack. The artifact of this skill is failing tests,\n  not a prose list.\n- **Validation in the wrong place.** Re-validating inside already-typed internal\n  code while the actual boundary (the `serde` call, the `abi.decode`) goes\n  unchecked.\n- **Passing \"abuse\" tests.** A reproduction that constructs invalid state and\n  asserts it is rejected proves only that invalid things are invalid. Exercise\n  the REAL code path with realistic fixtures so the test fails until the real\n  fix lands.\n- **Scope creep into the post-hoc review.** This is design-time. Do not turn it\n  into a line-by-line audit of a finished diff — that is the separate\n  security-review pass.\n\n## Verification\n\nThe work is done only when:\n\n- The trust boundaries, assets, and STRIDE notes for this change are written\n  down.\n- Each credible abuse case has a `forge test` or `cargo test` that FAILED before\n  the fix and PASSES after — show the before\u002Fafter run, not an assertion that it\n  \"should\" pass.\n- `cargo audit` was run and any new dependency's build script was read.\n- `git diff --cached` confirms no secret, key, or raw amount is being committed.\n- For onchain changes: a reentrancy \u002F unauthorized-caller \u002F rounding \u002F replay\n  test exists and passes; checks-effects-interactions verified.\n- For LLM tooling: model output is handled as data and the tool's\n  `allowed-tools` are granular.\n\n\"Seems secure\" is never acceptable. The evidence is the failing-then-passing\nabuse tests plus the boundary map.\n",{"name":5,"description":7,"allowed-tools":43,"user-invocable":54},[44,45,46,47,48,49,50,51,52,53],"Read","Grep","Glob","Write","Edit","Bash(rg *)","Bash(cargo test *)","Bash(forge test *)","Bash(cargo audit *)","Bash(git diff --cached *)",true,"master","5530d3fc73392f1e7dfe0b63cb40e662122fb79f","ai\u002Fskills\u002Fthreat-model-first",0,"https:\u002F\u002Fgithub.com\u002F0xgleb\u002Fdotconfig","https:\u002F\u002Fgithub.com\u002F0xgleb","2026-11-14 09:46:23",{"source":8,"skillName":5,"commands":63,"prompt":73,"zipAvailable":54},[64,67,70],{"tool":65,"command":66},"npx","npx skills add 0xgleb\u002Fdotconfig --skill threat-model-first",{"tool":68,"command":69},"bunx","bunx skills add 0xgleb\u002Fdotconfig --skill threat-model-first",{"tool":71,"command":72},"pnpm","pnpm dlx skills add 0xgleb\u002Fdotconfig --skill threat-model-first","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002F0xgleb\u002Fdotconfig\u002Fblob\u002Fmaster\u002Fai\u002Fskills\u002Fthreat-model-first\u002FSKILL.md",[],[76,100,125,151,168,194],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":38},"01M6DNKHP7NV7KR552KFZ6EW3R","worktree","Manage worktrees only for concurrent work by the owner or agents launched by the owner on the same repository, or for explicitly requested inventory and cleanup; never create one for isolation alone.","https:\u002F\u002Fgithub.com\u002F0xgleb\u002Fdotconfig\u002Fblob\u002Fmaster\u002Fai\u002Fskills\u002Fworktree\u002FSKILL.md",[],[78,83,84,85,86,87,88,89,90,91,92,93,94,31,95,96,97,98,99],"create","work","never","worktrees","manage","inventory","repository","fix","git","only","cleanup","cd","isolation","ls","for","mkdir","true",{"skillNo":101,"name":102,"nameZh":6,"description":103,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":104,"categories":105,"keywords":106,"publishedAt":38},"01M6DNKEGPR8F3BA95ZMWQ48V9","unslop","Rewrite draft prose before owner-facing replies, docs, issues, PR text, handoffs, or Telegram reports to remove AI tells while preserving facts, links, tone, and authority; auto-trigger on externally visible prose, but never on code, commands, logs, tool output, quoted sources, or user text.","https:\u002F\u002Fgithub.com\u002F0xgleb\u002Fdotconfig\u002Fblob\u002Fmaster\u002Fai\u002Fskills\u002Funslop\u002FSKILL.md",[],[102,107,85,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124],"draft","pr","remove","trigger","tool","issues","auto","rewrite","code","tone","telegram","sources","links","authority","logs","facts","output","replies",{"skillNo":126,"name":127,"nameZh":6,"description":128,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":129,"categories":130,"keywords":136,"publishedAt":38},"01M6DNK6M72QAMH2TYVMPYJGDM","test-inspector","Auto-run for new or changed retained behavior, even without test changes, and for test, fixture, snapshot or mock changes; verify real happy-path and error-path coverage, expose false-confidence assertions and over-mocking, but do not perform general production-code correctness review.","https:\u002F\u002Fgithub.com\u002F0xgleb\u002Fdotconfig\u002Fblob\u002Fmaster\u002Fai\u002Fskills\u002Ftest-inspector\u002FSKILL.md",[131],{"categoryNo":132,"categoryName":133,"taxonomyType":134,"taxonomyTypeDesc":135},"01JBSEEDCAT000000000000008","测试与质量","SCENARIO","使用场景",[137,138,139,140,113,141,31,115,29,142,26,143,144,145,146,147,148,91,149,150],"test","test inspector","fixture","do","false","run","coverage","error","number","URL","mktemp","snapshot","but","path",{"skillNo":152,"name":153,"nameZh":6,"description":154,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":155,"categories":156,"keywords":157,"publishedAt":38},"01M6DNK3BPN73D003FFADEWJP5","strong-typing-inspector","Lightweight review — flags primitives used where a domain type already exists (e.g. String for a USD field when Usd exists) and missed newtype opportunities.","https:\u002F\u002Fgithub.com\u002F0xgleb\u002Fdotconfig\u002Fblob\u002Fmaster\u002Fai\u002Fskills\u002Fstrong-typing-inspector\u002FSKILL.md",[],[158,159,137,160,108,161,26,162,163,91,164,165,31,166,29,167],"strong typing inspector","type","missed","domain","flags","grep","usd","gh","primitives","newtype",{"skillNo":169,"name":170,"nameZh":6,"description":171,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":172,"categories":173,"keywords":177,"publishedAt":38},"01M6DNJZHG2RK3MC4VPB9KWPYF","spec-review","Critique existing specifications and design proposals, including whether their approach or addition is justified, and refine them through adversarial Pi review and authorized author revisions. Not for code review, implementation-plan sequencing, creating a design from scratch or minor copy edits.","https:\u002F\u002Fgithub.com\u002F0xgleb\u002Fdotconfig\u002Fblob\u002Fmaster\u002Fai\u002Fskills\u002Fspec-review\u002FSKILL.md",[174],{"categoryNo":175,"categoryName":176,"taxonomyType":134,"taxonomyTypeDesc":135},"01JBSEEDCAT000000000000007","代码审查",[178,179,29,180,181,182,183,184,34,185,115,97,186,187,188,189,190,191,192,193],"spec review","spec","plan","critique","code review","pi","refine","implementation","sequencing","specifications","proposals","scratch","Existing","author","From Scratch","adversarial",{"skillNo":195,"name":196,"nameZh":6,"description":197,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":9,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":198,"categories":199,"keywords":200,"publishedAt":38},"01M6DNJVKEXDSYCM98DP1SRN0Y","shape-work","Use when a vague idea, issue, or potentially irreversible or multi-file change needs intent shaping before planning or an ADR; trigger on \"shape this idea\", \"explore the design space\", \"stress-test this direction\", \"lock intent\", or \"interview me\" regardless of implementation language, and never use it to plan or write code.","https:\u002F\u002Fgithub.com\u002F0xgleb\u002Fdotconfig\u002Fblob\u002Fmaster\u002Fai\u002Fskills\u002Fshape-work\u002FSKILL.md",[],[84,201,202,203,204,205,85,206,207,110,208,34,165,209,210,31,211,115,212,137],"shape work","shape","gh issue","interview","adr","view","planning","explore","stress-test","use","file","shaping"]