[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-trail-of-bits-writing-great-skills":3,"mdc-3b6kdp-key":34,"related-repo-trail-of-bits-writing-great-skills":952,"related-org-trail-of-bits-writing-great-skills":1067},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":12,"stars":23,"repoUrl":24,"updatedAt":25,"license":26,"forks":27,"topics":28,"repo":29,"sourceUrl":32,"mdContent":33},"writing-great-skills","author and edit agent skills","Reference for writing and editing agent skills well — the vocabulary and principles that make a skill predictable. Consult when authoring, reviewing, or pruning a SKILL.md.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},"trail-of-bits","Trail of Bits","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Ftrail-of-bits.png","trailofbits",[13,17,20],{"name":14,"slug":15,"type":16},"Documentation","documentation","tag",{"name":18,"slug":19,"type":16},"Best Practices","best-practices",{"name":21,"slug":22,"type":16},"Technical Writing","technical-writing",460,"https:\u002F\u002Fgithub.com\u002Ftrailofbits\u002Fskills-curated","2026-07-18T05:47:02.742938",null,29,[],{"repoUrl":24,"stars":23,"forks":27,"topics":30,"description":31},[],"Curated, community-vetted Claude Code plugin marketplace","https:\u002F\u002Fgithub.com\u002Ftrailofbits\u002Fskills-curated\u002Ftree\u002FHEAD\u002Fplugins\u002Fproductivity\u002Fskills\u002Fwriting-great-skills","---\nname: writing-great-skills\ndescription: >-\n  Reference for writing and editing agent skills well — the vocabulary and\n  principles that make a skill predictable. Consult when authoring, reviewing,\n  or pruning a SKILL.md.\ndisable-model-invocation: true\nallowed-tools:\n  - Read\n  - Grep\n  - Glob\n  - Edit\n  - Write\n---\n\n# Writing Great Skills\n\nA skill exists to wrangle determinism out of a stochastic system. **Predictability** — the agent taking the same _process_ every run, not producing the same output — is the root virtue; every lever below serves it.\n\n**Bold terms** are defined in [`GLOSSARY.md`]({baseDir}\u002Freferences\u002FGLOSSARY.md); look them up there for the full meaning.\n\n## When to Use\n\n- Writing a new skill and deciding how to structure, word, and scope it\n- Reviewing or editing an existing skill that fires unreliably, runs\n  inconsistently, or has grown bloated\n- Diagnosing a skill failure mode: the agent rushes steps, ignores reference\n  material, or triggers the wrong skill\n\n## When NOT to Use\n\n- General prompt engineering for one-off prompts — this reference is about\n  reusable skills, where predictability across runs is the goal\n- Questions about skill\u002Fplugin packaging mechanics (directory layout,\n  manifest fields); consult the platform's plugin documentation instead\n\n## Invocation\n\nTwo choices, trading different costs:\n\n- A **model-invoked** skill keeps a **description**, so the agent can fire it autonomously _and_ other skills can reach it (you can still type its name too). It contributes to **context load** — the description sits in the window every turn. Mechanics: omit `disable-model-invocation`, and write a model-facing description with rich trigger phrasing (\"Use when the user wants…, mentions…\").\n- A **user-invoked** skill strips the description from the agent's reach: only you, typing its name, can invoke it — and no other skill can. Zero context load, but it spends **cognitive load**: _you_ are the index that must remember it exists. Mechanics: set `disable-model-invocation: true`; the `description` becomes human-facing — a one-line summary, trigger lists stripped.\n\nPick model-invocation only when the agent must reach the skill on its own, or another skill must. If it only ever fires by hand, make it user-invoked and pay no context load.\n\nWhen user-invoked skills multiply past what you can remember, that piled-up cognitive load is cured by a **router skill**: one user-invoked skill that names the others and when to reach for each.\n\n## Writing the description\n\nA model-invoked **description** does two jobs — state what the skill is, and list the **branches** that should trigger it. Every word increases **context load**, so a description earns even harder pruning than the body:\n\n- **Front-load the skill's leading word** — the description is where it does its invocation work.\n- **One trigger per branch.** Synonyms that rename a single branch are **duplication** — \"build features using TDD … asks for test-first development\" is one branch written twice. Collapse them; keep only genuinely distinct branches.\n- **Cut identity that's already in the body.** Keep the description to triggers, plus any \"when another skill needs…\" reach clause.\n\n## Information hierarchy\n\nA skill is built from two content types — **steps** and **reference** — that mix freely: a skill can be all steps, all reference, or both. The core decision is which to use and where each sits on the **information hierarchy**, a ladder ranked by how immediately the agent needs the material:\n\n1. **In-skill step** — an ordered action in `SKILL.md`, the primary tier: what the agent does, in order. Each step ends on a **completion criterion**, the condition that tells the agent the work is done. Make it _checkable_ (can the agent tell done from not-done?) and, where it matters, _exhaustive_ (\"every modified model accounted for\", not \"produce a change list\") — a vague criterion invites **premature completion**.\n2. **In-skill reference** — a definition, rule, or fact in `SKILL.md`, consulted on demand. Often a legitimately flat peer-set (every rule of a review on one rung) — a fine arrangement, not a smell. _This skill is all reference._\n3. **External reference** — reference pushed out of `SKILL.md` into a separate file, reached by a **context pointer**, loaded only when the pointer fires. (Spans _disclosed_ reference — a sibling file like `GLOSSARY.md`, still part of the skill — through fully **external reference** that lives outside the skill system and any skill can point at.)\n\nA demanding completion criterion drives thorough **legwork** — the digging the agent does within the work — whether the skill has steps or not, since \"every rule applied\" binds flat reference just as \"every step done\" binds a sequence.\n\nPush too little down and the top bloats; push too much and you hide material the agent actually needs. That tension is the whole decision.\n\n**Progressive disclosure** is the move down the ladder — out of `SKILL.md` into a linked file — so the top stays legible. Mechanics: a linked `.md` file in the skill folder, named for what it holds (this skill discloses its full definitions to `GLOSSARY.md`). Some skills are used in more than one way, and each distinct way is a **branch** — different runs taking different paths through the skill. Branching is the cleanest disclosure test: inline what every branch needs, and push behind a pointer what only some branches reach. A **context pointer**'s _wording_, not its target, decides when and how reliably the agent reaches the material.\n\nWhere the ladder decides _how far down_ a piece sits, **co-location** decides _what sits beside it_ once there: keep a concept's definition, rules, and caveats under one heading rather than scattered, so reading one part brings its neighbours with it.\n\n## When to split\n\n**Granularity** is how finely you divide skills, and each cut spends one of the two loads, so split only when the cut earns it. Two cuts:\n\n- **By invocation** — split off a **model-invoked** skill when you have a distinct **leading word** that should trigger it on its own, or another skill must reach it. You pay **context load** for the new always-loaded **description**, so that independent reach has to be worth it.\n- **By sequence** — split a run of **steps** when the steps still ahead (a step's **post-completion steps**) tempt the agent to rush the one in front of it (**premature completion**). Keeping them out of view encourages the agent to do more **legwork** on the current task.\n\n## Pruning\n\nKeep each meaning in a **single source of truth**: one authoritative place, so changing the behaviour is a one-place edit.\n\nCheck every line for **relevance**: does it still bear on what the skill does?\n\nThen hunt **no-ops** sentence by sentence, not just line by line: run the no-op test on each sentence in isolation, and when one fails, delete the whole sentence rather than trim words from it. Be aggressive — most prose that fails should go, not be rewritten.\n\n## Leading words\n\nA **leading word** is a compact concept already living in the model's pretraining that the agent thinks with while running the skill (e.g. _lesson_, _fog of war_, _tracer bullets_). Repeated throughout the text (though not necessarily - a strong leading word might only be needed once), it accumulates a distributed definition and anchors a whole region of behaviour in the fewest tokens, by recruiting priors the model already holds.\n\nIt serves predictability twice. In the body it anchors _execution_: the agent reaches for the same behaviour every time the word appears. In the description it anchors _invocation_: when the same word lives in your prompts, docs, and code, the agent links that shared language to the skill and fires it more reliably.\n\nHunt for opportunities to refactor skills to use leading words. A triad spelled out at three sites (**duplication**), a description spending a sentence to gesture at one idea — each is a passage begging to **collapse** into a single token. Examples include:\n\n- \"fast, deterministic, low-overhead\" -> _tight_ — one quality restated across a phase — into a single pretrained word (a _tight_ loop).\n- \"a loop you believe in\" -> _red_ — converts a fuzzy gate into a binary observable state (the loop goes _red_ on the bug, or it doesn't).\n\nYou win twice over: fewer tokens, _and_ a sharper hook for the agent to hang its thinking on. Assume every skill is carrying restatements that leading words retire — go find them.\n\n## Failure modes\n\nUse these to diagnose issues the user may be having with the skill.\n\n- **Premature completion** — ending a step before it's genuinely done, attention slipping to _being done_. Defence, in order: sharpen the completion criterion first (cheap, local); only if it is irreducibly fuzzy _and_ you observe the rush, hide the post-completion steps by splitting (the sequence cut).\n- **Duplication** — the same meaning in more than one place. Costs maintenance and tokens, and inflates a meaning's prominence on the ladder past its real rank.\n- **Sediment** — stale layers that settle because adding feels safe and removing feels risky. The default fate of any skill without a pruning discipline.\n- **Sprawl** — a skill simply too long, even when every line is live and unique. Hurts readability and maintainability and wastes tokens. The cure is the ladder: disclose **reference** behind pointers, and split by **branch** or sequence so each path carries only what it needs.\n- **No-op** — a line the model already obeys by default, so you pay load to say nothing. The test: does it change behaviour versus the default? A weak leading word (_be thorough_ when the agent is already thorough-ish) is a no-op; the fix is a stronger word (_relentless_), not a different technique.\n- **Negation** — steering by prohibition backfires: _don't think of an elephant_ names the elephant and makes it more available, not less. Prompt the **positive** — state the target behaviour so the banned one is never spoken; keep a prohibition only as a hard guardrail you can't phrase positively, and even then pair it with what to do instead.\n",{"data":35,"body":43},{"name":4,"description":6,"disable-model-invocation":36,"allowed-tools":37},true,[38,39,40,41,42],"Read","Grep","Glob","Edit","Write",{"type":44,"children":45},"root",[46,54,76,100,107,127,133,146,152,157,241,246,258,264,288,328,334,360,477,489,494,546,572,578,588,661,667,679,691,703,709,739,757,775,814,825,831,836],{"type":47,"tag":48,"props":49,"children":50},"element","h1",{"id":4},[51],{"type":52,"value":53},"text","Writing Great Skills",{"type":47,"tag":55,"props":56,"children":57},"p",{},[58,60,66,68,74],{"type":52,"value":59},"A skill exists to wrangle determinism out of a stochastic system. ",{"type":47,"tag":61,"props":62,"children":63},"strong",{},[64],{"type":52,"value":65},"Predictability",{"type":52,"value":67}," — the agent taking the same ",{"type":47,"tag":69,"props":70,"children":71},"em",{},[72],{"type":52,"value":73},"process",{"type":52,"value":75}," every run, not producing the same output — is the root virtue; every lever below serves it.",{"type":47,"tag":55,"props":77,"children":78},{},[79,84,86,98],{"type":47,"tag":61,"props":80,"children":81},{},[82],{"type":52,"value":83},"Bold terms",{"type":52,"value":85}," are defined in ",{"type":47,"tag":87,"props":88,"children":90},"a",{"href":89},"%7BbaseDir%7D\u002Freferences\u002FGLOSSARY.md",[91],{"type":47,"tag":92,"props":93,"children":95},"code",{"className":94},[],[96],{"type":52,"value":97},"GLOSSARY.md",{"type":52,"value":99},"; look them up there for the full meaning.",{"type":47,"tag":101,"props":102,"children":104},"h2",{"id":103},"when-to-use",[105],{"type":52,"value":106},"When to Use",{"type":47,"tag":108,"props":109,"children":110},"ul",{},[111,117,122],{"type":47,"tag":112,"props":113,"children":114},"li",{},[115],{"type":52,"value":116},"Writing a new skill and deciding how to structure, word, and scope it",{"type":47,"tag":112,"props":118,"children":119},{},[120],{"type":52,"value":121},"Reviewing or editing an existing skill that fires unreliably, runs\ninconsistently, or has grown bloated",{"type":47,"tag":112,"props":123,"children":124},{},[125],{"type":52,"value":126},"Diagnosing a skill failure mode: the agent rushes steps, ignores reference\nmaterial, or triggers the wrong skill",{"type":47,"tag":101,"props":128,"children":130},{"id":129},"when-not-to-use",[131],{"type":52,"value":132},"When NOT to Use",{"type":47,"tag":108,"props":134,"children":135},{},[136,141],{"type":47,"tag":112,"props":137,"children":138},{},[139],{"type":52,"value":140},"General prompt engineering for one-off prompts — this reference is about\nreusable skills, where predictability across runs is the goal",{"type":47,"tag":112,"props":142,"children":143},{},[144],{"type":52,"value":145},"Questions about skill\u002Fplugin packaging mechanics (directory layout,\nmanifest fields); consult the platform's plugin documentation instead",{"type":47,"tag":101,"props":147,"children":149},{"id":148},"invocation",[150],{"type":52,"value":151},"Invocation",{"type":47,"tag":55,"props":153,"children":154},{},[155],{"type":52,"value":156},"Two choices, trading different costs:",{"type":47,"tag":108,"props":158,"children":159},{},[160,201],{"type":47,"tag":112,"props":161,"children":162},{},[163,165,170,172,177,179,184,186,191,193,199],{"type":52,"value":164},"A ",{"type":47,"tag":61,"props":166,"children":167},{},[168],{"type":52,"value":169},"model-invoked",{"type":52,"value":171}," skill keeps a ",{"type":47,"tag":61,"props":173,"children":174},{},[175],{"type":52,"value":176},"description",{"type":52,"value":178},", so the agent can fire it autonomously ",{"type":47,"tag":69,"props":180,"children":181},{},[182],{"type":52,"value":183},"and",{"type":52,"value":185}," other skills can reach it (you can still type its name too). It contributes to ",{"type":47,"tag":61,"props":187,"children":188},{},[189],{"type":52,"value":190},"context load",{"type":52,"value":192}," — the description sits in the window every turn. Mechanics: omit ",{"type":47,"tag":92,"props":194,"children":196},{"className":195},[],[197],{"type":52,"value":198},"disable-model-invocation",{"type":52,"value":200},", and write a model-facing description with rich trigger phrasing (\"Use when the user wants…, mentions…\").",{"type":47,"tag":112,"props":202,"children":203},{},[204,205,210,212,217,219,224,226,232,234,239],{"type":52,"value":164},{"type":47,"tag":61,"props":206,"children":207},{},[208],{"type":52,"value":209},"user-invoked",{"type":52,"value":211}," skill strips the description from the agent's reach: only you, typing its name, can invoke it — and no other skill can. Zero context load, but it spends ",{"type":47,"tag":61,"props":213,"children":214},{},[215],{"type":52,"value":216},"cognitive load",{"type":52,"value":218},": ",{"type":47,"tag":69,"props":220,"children":221},{},[222],{"type":52,"value":223},"you",{"type":52,"value":225}," are the index that must remember it exists. Mechanics: set ",{"type":47,"tag":92,"props":227,"children":229},{"className":228},[],[230],{"type":52,"value":231},"disable-model-invocation: true",{"type":52,"value":233},"; the ",{"type":47,"tag":92,"props":235,"children":237},{"className":236},[],[238],{"type":52,"value":176},{"type":52,"value":240}," becomes human-facing — a one-line summary, trigger lists stripped.",{"type":47,"tag":55,"props":242,"children":243},{},[244],{"type":52,"value":245},"Pick model-invocation only when the agent must reach the skill on its own, or another skill must. If it only ever fires by hand, make it user-invoked and pay no context load.",{"type":47,"tag":55,"props":247,"children":248},{},[249,251,256],{"type":52,"value":250},"When user-invoked skills multiply past what you can remember, that piled-up cognitive load is cured by a ",{"type":47,"tag":61,"props":252,"children":253},{},[254],{"type":52,"value":255},"router skill",{"type":52,"value":257},": one user-invoked skill that names the others and when to reach for each.",{"type":47,"tag":101,"props":259,"children":261},{"id":260},"writing-the-description",[262],{"type":52,"value":263},"Writing the description",{"type":47,"tag":55,"props":265,"children":266},{},[267,269,273,275,280,282,286],{"type":52,"value":268},"A model-invoked ",{"type":47,"tag":61,"props":270,"children":271},{},[272],{"type":52,"value":176},{"type":52,"value":274}," does two jobs — state what the skill is, and list the ",{"type":47,"tag":61,"props":276,"children":277},{},[278],{"type":52,"value":279},"branches",{"type":52,"value":281}," that should trigger it. Every word increases ",{"type":47,"tag":61,"props":283,"children":284},{},[285],{"type":52,"value":190},{"type":52,"value":287},", so a description earns even harder pruning than the body:",{"type":47,"tag":108,"props":289,"children":290},{},[291,301,318],{"type":47,"tag":112,"props":292,"children":293},{},[294,299],{"type":47,"tag":61,"props":295,"children":296},{},[297],{"type":52,"value":298},"Front-load the skill's leading word",{"type":52,"value":300}," — the description is where it does its invocation work.",{"type":47,"tag":112,"props":302,"children":303},{},[304,309,311,316],{"type":47,"tag":61,"props":305,"children":306},{},[307],{"type":52,"value":308},"One trigger per branch.",{"type":52,"value":310}," Synonyms that rename a single branch are ",{"type":47,"tag":61,"props":312,"children":313},{},[314],{"type":52,"value":315},"duplication",{"type":52,"value":317}," — \"build features using TDD … asks for test-first development\" is one branch written twice. Collapse them; keep only genuinely distinct branches.",{"type":47,"tag":112,"props":319,"children":320},{},[321,326],{"type":47,"tag":61,"props":322,"children":323},{},[324],{"type":52,"value":325},"Cut identity that's already in the body.",{"type":52,"value":327}," Keep the description to triggers, plus any \"when another skill needs…\" reach clause.",{"type":47,"tag":101,"props":329,"children":331},{"id":330},"information-hierarchy",[332],{"type":52,"value":333},"Information hierarchy",{"type":47,"tag":55,"props":335,"children":336},{},[337,339,344,346,351,353,358],{"type":52,"value":338},"A skill is built from two content types — ",{"type":47,"tag":61,"props":340,"children":341},{},[342],{"type":52,"value":343},"steps",{"type":52,"value":345}," and ",{"type":47,"tag":61,"props":347,"children":348},{},[349],{"type":52,"value":350},"reference",{"type":52,"value":352}," — that mix freely: a skill can be all steps, all reference, or both. The core decision is which to use and where each sits on the ",{"type":47,"tag":61,"props":354,"children":355},{},[356],{"type":52,"value":357},"information hierarchy",{"type":52,"value":359},", a ladder ranked by how immediately the agent needs the material:",{"type":47,"tag":361,"props":362,"children":363},"ol",{},[364,410,432],{"type":47,"tag":112,"props":365,"children":366},{},[367,372,374,380,382,387,389,394,396,401,403,408],{"type":47,"tag":61,"props":368,"children":369},{},[370],{"type":52,"value":371},"In-skill step",{"type":52,"value":373}," — an ordered action in ",{"type":47,"tag":92,"props":375,"children":377},{"className":376},[],[378],{"type":52,"value":379},"SKILL.md",{"type":52,"value":381},", the primary tier: what the agent does, in order. Each step ends on a ",{"type":47,"tag":61,"props":383,"children":384},{},[385],{"type":52,"value":386},"completion criterion",{"type":52,"value":388},", the condition that tells the agent the work is done. Make it ",{"type":47,"tag":69,"props":390,"children":391},{},[392],{"type":52,"value":393},"checkable",{"type":52,"value":395}," (can the agent tell done from not-done?) and, where it matters, ",{"type":47,"tag":69,"props":397,"children":398},{},[399],{"type":52,"value":400},"exhaustive",{"type":52,"value":402}," (\"every modified model accounted for\", not \"produce a change list\") — a vague criterion invites ",{"type":47,"tag":61,"props":404,"children":405},{},[406],{"type":52,"value":407},"premature completion",{"type":52,"value":409},".",{"type":47,"tag":112,"props":411,"children":412},{},[413,418,420,425,427],{"type":47,"tag":61,"props":414,"children":415},{},[416],{"type":52,"value":417},"In-skill reference",{"type":52,"value":419}," — a definition, rule, or fact in ",{"type":47,"tag":92,"props":421,"children":423},{"className":422},[],[424],{"type":52,"value":379},{"type":52,"value":426},", consulted on demand. Often a legitimately flat peer-set (every rule of a review on one rung) — a fine arrangement, not a smell. ",{"type":47,"tag":69,"props":428,"children":429},{},[430],{"type":52,"value":431},"This skill is all reference.",{"type":47,"tag":112,"props":433,"children":434},{},[435,440,442,447,449,454,456,461,463,468,470,475],{"type":47,"tag":61,"props":436,"children":437},{},[438],{"type":52,"value":439},"External reference",{"type":52,"value":441}," — reference pushed out of ",{"type":47,"tag":92,"props":443,"children":445},{"className":444},[],[446],{"type":52,"value":379},{"type":52,"value":448}," into a separate file, reached by a ",{"type":47,"tag":61,"props":450,"children":451},{},[452],{"type":52,"value":453},"context pointer",{"type":52,"value":455},", loaded only when the pointer fires. (Spans ",{"type":47,"tag":69,"props":457,"children":458},{},[459],{"type":52,"value":460},"disclosed",{"type":52,"value":462}," reference — a sibling file like ",{"type":47,"tag":92,"props":464,"children":466},{"className":465},[],[467],{"type":52,"value":97},{"type":52,"value":469},", still part of the skill — through fully ",{"type":47,"tag":61,"props":471,"children":472},{},[473],{"type":52,"value":474},"external reference",{"type":52,"value":476}," that lives outside the skill system and any skill can point at.)",{"type":47,"tag":55,"props":478,"children":479},{},[480,482,487],{"type":52,"value":481},"A demanding completion criterion drives thorough ",{"type":47,"tag":61,"props":483,"children":484},{},[485],{"type":52,"value":486},"legwork",{"type":52,"value":488}," — the digging the agent does within the work — whether the skill has steps or not, since \"every rule applied\" binds flat reference just as \"every step done\" binds a sequence.",{"type":47,"tag":55,"props":490,"children":491},{},[492],{"type":52,"value":493},"Push too little down and the top bloats; push too much and you hide material the agent actually needs. That tension is the whole decision.",{"type":47,"tag":55,"props":495,"children":496},{},[497,502,504,509,511,517,519,524,526,531,533,537,539,544],{"type":47,"tag":61,"props":498,"children":499},{},[500],{"type":52,"value":501},"Progressive disclosure",{"type":52,"value":503}," is the move down the ladder — out of ",{"type":47,"tag":92,"props":505,"children":507},{"className":506},[],[508],{"type":52,"value":379},{"type":52,"value":510}," into a linked file — so the top stays legible. Mechanics: a linked ",{"type":47,"tag":92,"props":512,"children":514},{"className":513},[],[515],{"type":52,"value":516},".md",{"type":52,"value":518}," file in the skill folder, named for what it holds (this skill discloses its full definitions to ",{"type":47,"tag":92,"props":520,"children":522},{"className":521},[],[523],{"type":52,"value":97},{"type":52,"value":525},"). Some skills are used in more than one way, and each distinct way is a ",{"type":47,"tag":61,"props":527,"children":528},{},[529],{"type":52,"value":530},"branch",{"type":52,"value":532}," — different runs taking different paths through the skill. Branching is the cleanest disclosure test: inline what every branch needs, and push behind a pointer what only some branches reach. A ",{"type":47,"tag":61,"props":534,"children":535},{},[536],{"type":52,"value":453},{"type":52,"value":538},"'s ",{"type":47,"tag":69,"props":540,"children":541},{},[542],{"type":52,"value":543},"wording",{"type":52,"value":545},", not its target, decides when and how reliably the agent reaches the material.",{"type":47,"tag":55,"props":547,"children":548},{},[549,551,556,558,563,565,570],{"type":52,"value":550},"Where the ladder decides ",{"type":47,"tag":69,"props":552,"children":553},{},[554],{"type":52,"value":555},"how far down",{"type":52,"value":557}," a piece sits, ",{"type":47,"tag":61,"props":559,"children":560},{},[561],{"type":52,"value":562},"co-location",{"type":52,"value":564}," decides ",{"type":47,"tag":69,"props":566,"children":567},{},[568],{"type":52,"value":569},"what sits beside it",{"type":52,"value":571}," once there: keep a concept's definition, rules, and caveats under one heading rather than scattered, so reading one part brings its neighbours with it.",{"type":47,"tag":101,"props":573,"children":575},{"id":574},"when-to-split",[576],{"type":52,"value":577},"When to split",{"type":47,"tag":55,"props":579,"children":580},{},[581,586],{"type":47,"tag":61,"props":582,"children":583},{},[584],{"type":52,"value":585},"Granularity",{"type":52,"value":587}," is how finely you divide skills, and each cut spends one of the two loads, so split only when the cut earns it. Two cuts:",{"type":47,"tag":108,"props":589,"children":590},{},[591,626],{"type":47,"tag":112,"props":592,"children":593},{},[594,599,601,605,607,612,614,618,620,624],{"type":47,"tag":61,"props":595,"children":596},{},[597],{"type":52,"value":598},"By invocation",{"type":52,"value":600}," — split off a ",{"type":47,"tag":61,"props":602,"children":603},{},[604],{"type":52,"value":169},{"type":52,"value":606}," skill when you have a distinct ",{"type":47,"tag":61,"props":608,"children":609},{},[610],{"type":52,"value":611},"leading word",{"type":52,"value":613}," that should trigger it on its own, or another skill must reach it. You pay ",{"type":47,"tag":61,"props":615,"children":616},{},[617],{"type":52,"value":190},{"type":52,"value":619}," for the new always-loaded ",{"type":47,"tag":61,"props":621,"children":622},{},[623],{"type":52,"value":176},{"type":52,"value":625},", so that independent reach has to be worth it.",{"type":47,"tag":112,"props":627,"children":628},{},[629,634,636,640,642,647,649,653,655,659],{"type":47,"tag":61,"props":630,"children":631},{},[632],{"type":52,"value":633},"By sequence",{"type":52,"value":635}," — split a run of ",{"type":47,"tag":61,"props":637,"children":638},{},[639],{"type":52,"value":343},{"type":52,"value":641}," when the steps still ahead (a step's ",{"type":47,"tag":61,"props":643,"children":644},{},[645],{"type":52,"value":646},"post-completion steps",{"type":52,"value":648},") tempt the agent to rush the one in front of it (",{"type":47,"tag":61,"props":650,"children":651},{},[652],{"type":52,"value":407},{"type":52,"value":654},"). Keeping them out of view encourages the agent to do more ",{"type":47,"tag":61,"props":656,"children":657},{},[658],{"type":52,"value":486},{"type":52,"value":660}," on the current task.",{"type":47,"tag":101,"props":662,"children":664},{"id":663},"pruning",[665],{"type":52,"value":666},"Pruning",{"type":47,"tag":55,"props":668,"children":669},{},[670,672,677],{"type":52,"value":671},"Keep each meaning in a ",{"type":47,"tag":61,"props":673,"children":674},{},[675],{"type":52,"value":676},"single source of truth",{"type":52,"value":678},": one authoritative place, so changing the behaviour is a one-place edit.",{"type":47,"tag":55,"props":680,"children":681},{},[682,684,689],{"type":52,"value":683},"Check every line for ",{"type":47,"tag":61,"props":685,"children":686},{},[687],{"type":52,"value":688},"relevance",{"type":52,"value":690},": does it still bear on what the skill does?",{"type":47,"tag":55,"props":692,"children":693},{},[694,696,701],{"type":52,"value":695},"Then hunt ",{"type":47,"tag":61,"props":697,"children":698},{},[699],{"type":52,"value":700},"no-ops",{"type":52,"value":702}," sentence by sentence, not just line by line: run the no-op test on each sentence in isolation, and when one fails, delete the whole sentence rather than trim words from it. Be aggressive — most prose that fails should go, not be rewritten.",{"type":47,"tag":101,"props":704,"children":706},{"id":705},"leading-words",[707],{"type":52,"value":708},"Leading words",{"type":47,"tag":55,"props":710,"children":711},{},[712,713,717,719,724,726,731,732,737],{"type":52,"value":164},{"type":47,"tag":61,"props":714,"children":715},{},[716],{"type":52,"value":611},{"type":52,"value":718}," is a compact concept already living in the model's pretraining that the agent thinks with while running the skill (e.g. ",{"type":47,"tag":69,"props":720,"children":721},{},[722],{"type":52,"value":723},"lesson",{"type":52,"value":725},", ",{"type":47,"tag":69,"props":727,"children":728},{},[729],{"type":52,"value":730},"fog of war",{"type":52,"value":725},{"type":47,"tag":69,"props":733,"children":734},{},[735],{"type":52,"value":736},"tracer bullets",{"type":52,"value":738},"). Repeated throughout the text (though not necessarily - a strong leading word might only be needed once), it accumulates a distributed definition and anchors a whole region of behaviour in the fewest tokens, by recruiting priors the model already holds.",{"type":47,"tag":55,"props":740,"children":741},{},[742,744,749,751,755],{"type":52,"value":743},"It serves predictability twice. In the body it anchors ",{"type":47,"tag":69,"props":745,"children":746},{},[747],{"type":52,"value":748},"execution",{"type":52,"value":750},": the agent reaches for the same behaviour every time the word appears. In the description it anchors ",{"type":47,"tag":69,"props":752,"children":753},{},[754],{"type":52,"value":148},{"type":52,"value":756},": when the same word lives in your prompts, docs, and code, the agent links that shared language to the skill and fires it more reliably.",{"type":47,"tag":55,"props":758,"children":759},{},[760,762,766,768,773],{"type":52,"value":761},"Hunt for opportunities to refactor skills to use leading words. A triad spelled out at three sites (",{"type":47,"tag":61,"props":763,"children":764},{},[765],{"type":52,"value":315},{"type":52,"value":767},"), a description spending a sentence to gesture at one idea — each is a passage begging to ",{"type":47,"tag":61,"props":769,"children":770},{},[771],{"type":52,"value":772},"collapse",{"type":52,"value":774}," into a single token. Examples include:",{"type":47,"tag":108,"props":776,"children":777},{},[778,796],{"type":47,"tag":112,"props":779,"children":780},{},[781,783,788,790,794],{"type":52,"value":782},"\"fast, deterministic, low-overhead\" -> ",{"type":47,"tag":69,"props":784,"children":785},{},[786],{"type":52,"value":787},"tight",{"type":52,"value":789}," — one quality restated across a phase — into a single pretrained word (a ",{"type":47,"tag":69,"props":791,"children":792},{},[793],{"type":52,"value":787},{"type":52,"value":795}," loop).",{"type":47,"tag":112,"props":797,"children":798},{},[799,801,806,808,812],{"type":52,"value":800},"\"a loop you believe in\" -> ",{"type":47,"tag":69,"props":802,"children":803},{},[804],{"type":52,"value":805},"red",{"type":52,"value":807}," — converts a fuzzy gate into a binary observable state (the loop goes ",{"type":47,"tag":69,"props":809,"children":810},{},[811],{"type":52,"value":805},{"type":52,"value":813}," on the bug, or it doesn't).",{"type":47,"tag":55,"props":815,"children":816},{},[817,819,823],{"type":52,"value":818},"You win twice over: fewer tokens, ",{"type":47,"tag":69,"props":820,"children":821},{},[822],{"type":52,"value":183},{"type":52,"value":824}," a sharper hook for the agent to hang its thinking on. Assume every skill is carrying restatements that leading words retire — go find them.",{"type":47,"tag":101,"props":826,"children":828},{"id":827},"failure-modes",[829],{"type":52,"value":830},"Failure modes",{"type":47,"tag":55,"props":832,"children":833},{},[834],{"type":52,"value":835},"Use these to diagnose issues the user may be having with the skill.",{"type":47,"tag":108,"props":837,"children":838},{},[839,862,872,882,904,928],{"type":47,"tag":112,"props":840,"children":841},{},[842,847,849,854,856,860],{"type":47,"tag":61,"props":843,"children":844},{},[845],{"type":52,"value":846},"Premature completion",{"type":52,"value":848}," — ending a step before it's genuinely done, attention slipping to ",{"type":47,"tag":69,"props":850,"children":851},{},[852],{"type":52,"value":853},"being done",{"type":52,"value":855},". Defence, in order: sharpen the completion criterion first (cheap, local); only if it is irreducibly fuzzy ",{"type":47,"tag":69,"props":857,"children":858},{},[859],{"type":52,"value":183},{"type":52,"value":861}," you observe the rush, hide the post-completion steps by splitting (the sequence cut).",{"type":47,"tag":112,"props":863,"children":864},{},[865,870],{"type":47,"tag":61,"props":866,"children":867},{},[868],{"type":52,"value":869},"Duplication",{"type":52,"value":871}," — the same meaning in more than one place. Costs maintenance and tokens, and inflates a meaning's prominence on the ladder past its real rank.",{"type":47,"tag":112,"props":873,"children":874},{},[875,880],{"type":47,"tag":61,"props":876,"children":877},{},[878],{"type":52,"value":879},"Sediment",{"type":52,"value":881}," — stale layers that settle because adding feels safe and removing feels risky. The default fate of any skill without a pruning discipline.",{"type":47,"tag":112,"props":883,"children":884},{},[885,890,892,896,898,902],{"type":47,"tag":61,"props":886,"children":887},{},[888],{"type":52,"value":889},"Sprawl",{"type":52,"value":891}," — a skill simply too long, even when every line is live and unique. Hurts readability and maintainability and wastes tokens. The cure is the ladder: disclose ",{"type":47,"tag":61,"props":893,"children":894},{},[895],{"type":52,"value":350},{"type":52,"value":897}," behind pointers, and split by ",{"type":47,"tag":61,"props":899,"children":900},{},[901],{"type":52,"value":530},{"type":52,"value":903}," or sequence so each path carries only what it needs.",{"type":47,"tag":112,"props":905,"children":906},{},[907,912,914,919,921,926],{"type":47,"tag":61,"props":908,"children":909},{},[910],{"type":52,"value":911},"No-op",{"type":52,"value":913}," — a line the model already obeys by default, so you pay load to say nothing. The test: does it change behaviour versus the default? A weak leading word (",{"type":47,"tag":69,"props":915,"children":916},{},[917],{"type":52,"value":918},"be thorough",{"type":52,"value":920}," when the agent is already thorough-ish) is a no-op; the fix is a stronger word (",{"type":47,"tag":69,"props":922,"children":923},{},[924],{"type":52,"value":925},"relentless",{"type":52,"value":927},"), not a different technique.",{"type":47,"tag":112,"props":929,"children":930},{},[931,936,938,943,945,950],{"type":47,"tag":61,"props":932,"children":933},{},[934],{"type":52,"value":935},"Negation",{"type":52,"value":937}," — steering by prohibition backfires: ",{"type":47,"tag":69,"props":939,"children":940},{},[941],{"type":52,"value":942},"don't think of an elephant",{"type":52,"value":944}," names the elephant and makes it more available, not less. Prompt the ",{"type":47,"tag":61,"props":946,"children":947},{},[948],{"type":52,"value":949},"positive",{"type":52,"value":951}," — state the target behaviour so the banned one is never spoken; keep a prohibition only as a hard guardrail you can't phrase positively, and even then pair it with what to do instead.",{"items":953,"total":1066},[954,970,982,1001,1017,1033,1047],{"slug":955,"name":955,"fn":956,"description":957,"org":958,"tags":959,"stars":23,"repoUrl":24,"updatedAt":969},"ffuf-web-fuzzing","perform web fuzzing with ffuf","Expert guidance for ffuf web fuzzing during authorized penetration testing. Covers directory discovery, subdomain enumeration, parameter fuzzing, authenticated fuzzing with raw requests, auto-calibration, and result analysis. Use when running ffuf scans, analyzing ffuf output, or building fuzzing strategies for web targets.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[960,963,966],{"name":961,"slug":962,"type":16},"Code Analysis","code-analysis",{"name":964,"slug":965,"type":16},"Security","security",{"name":967,"slug":968,"type":16},"Testing","testing","2026-07-17T06:05:08.247908",{"slug":971,"name":971,"fn":972,"description":973,"org":974,"tags":975,"stars":23,"repoUrl":24,"updatedAt":981},"ghidra-headless","reverse engineer binaries with Ghidra","Reverse engineers binaries using Ghidra's headless analyzer. Use when decompiling executables, extracting functions, strings, symbols, or analyzing call graphs from compiled binaries without the Ghidra GUI.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[976,977,980],{"name":961,"slug":962,"type":16},{"name":978,"slug":979,"type":16},"Debugging","debugging",{"name":964,"slug":965,"type":16},"2026-07-18T05:47:30.015093",{"slug":983,"name":983,"fn":984,"description":985,"org":986,"tags":987,"stars":23,"repoUrl":24,"updatedAt":1000},"grilling","stress-test plans and decisions","Interviews the user relentlessly about a plan, decision, or idea until every branch of the decision tree is resolved. Use when the user wants to stress-test their thinking, sharpen a plan or design before acting, or uses any 'grill' trigger phrase (e.g. \"grill me on this\").",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[988,991,994,997],{"name":989,"slug":990,"type":16},"Analysis","analysis",{"name":992,"slug":993,"type":16},"Coaching","coaching",{"name":995,"slug":996,"type":16},"Ideation","ideation",{"name":998,"slug":999,"type":16},"Strategy","strategy","2026-07-18T05:48:12.46583",{"slug":1002,"name":1002,"fn":1003,"description":1004,"org":1005,"tags":1006,"stars":23,"repoUrl":24,"updatedAt":1016},"handoff","compact conversation for session handoff","Compacts the current conversation into a handoff document so a fresh agent can continue the work in a new session.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1007,1010,1013],{"name":1008,"slug":1009,"type":16},"Agents","agents",{"name":1011,"slug":1012,"type":16},"Context","context",{"name":1014,"slug":1015,"type":16},"Productivity","productivity","2026-07-18T05:47:03.196098",{"slug":1018,"name":1018,"fn":1019,"description":1020,"org":1021,"tags":1022,"stars":23,"repoUrl":24,"updatedAt":1032},"humanizer","edit text to sound human-written","Remove signs of AI-generated writing from text. Use when editing or reviewing\ntext to make it sound more natural and human-written. Based on Wikipedia's\ncomprehensive \"Signs of AI writing\" guide. Detects and fixes patterns including:\ninflated symbolism, promotional language, superficial -ing analyses, vague\nattributions, em dash overuse, rule of three, AI vocabulary words, negative\nparallelisms, and excessive conjunctive phrases. 30c5c8d (Update humanizer plugin to upstream v2.2.0)\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1023,1026,1029],{"name":1024,"slug":1025,"type":16},"Content Creation","content-creation",{"name":1027,"slug":1028,"type":16},"Editing","editing",{"name":1030,"slug":1031,"type":16},"Writing","writing","2026-07-18T05:47:18.1749",{"slug":1034,"name":1034,"fn":1035,"description":1036,"org":1037,"tags":1038,"stars":23,"repoUrl":24,"updatedAt":1046},"last30days","research recent community discussions and trends","Researches a topic from the last 30 days on Reddit, X, and the web. Surfaces real community discussions with engagement metrics and synthesizes findings into actionable insights. Use when the user wants to know what people are saying about a topic right now.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1039,1040,1043],{"name":989,"slug":990,"type":16},{"name":1041,"slug":1042,"type":16},"Research","research",{"name":1044,"slug":1045,"type":16},"Social Media","social-media","2026-07-17T06:04:39.744471",{"slug":1048,"name":1048,"fn":1049,"description":1050,"org":1051,"tags":1052,"stars":23,"repoUrl":24,"updatedAt":1065},"openai-cloudflare-deploy","deploy applications to Cloudflare","Deploy applications and infrastructure to Cloudflare using Workers, Pages, and related platform services. Use when the user asks to deploy, host, publish, or set up a project on Cloudflare. Originally from OpenAI's curated skills catalog.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1053,1056,1059,1062],{"name":1054,"slug":1055,"type":16},"Cloudflare","cloudflare",{"name":1057,"slug":1058,"type":16},"Cloudflare Pages","cloudflare-pages",{"name":1060,"slug":1061,"type":16},"Cloudflare Workers","cloudflare-workers",{"name":1063,"slug":1064,"type":16},"Deployment","deployment","2026-07-17T06:04:46.574433",31,{"items":1068,"total":1216},[1069,1084,1094,1110,1125,1136,1148,1158,1171,1182,1194,1205],{"slug":1070,"name":1070,"fn":1071,"description":1072,"org":1073,"tags":1074,"stars":1081,"repoUrl":1082,"updatedAt":1083},"address-sanitizer","detect memory errors during fuzzing","AddressSanitizer detects memory errors during fuzzing. Use when fuzzing C\u002FC++ code to find buffer overflows and use-after-free bugs.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1075,1078,1079,1080],{"name":1076,"slug":1077,"type":16},"C#","c",{"name":978,"slug":979,"type":16},{"name":964,"slug":965,"type":16},{"name":967,"slug":968,"type":16},6139,"https:\u002F\u002Fgithub.com\u002Ftrailofbits\u002Fskills","2026-07-17T06:05:14.925095",{"slug":1085,"name":1085,"fn":1086,"description":1087,"org":1088,"tags":1089,"stars":1081,"repoUrl":1082,"updatedAt":1093},"aflpp","perform multi-core fuzzing of C\u002FC++ projects","AFL++ is a fork of AFL with better fuzzing performance and advanced features. Use for multi-core fuzzing of C\u002FC++ projects.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1090,1091,1092],{"name":1076,"slug":1077,"type":16},{"name":964,"slug":965,"type":16},{"name":967,"slug":968,"type":16},"2026-07-17T06:05:12.433192",{"slug":1095,"name":1095,"fn":1096,"description":1097,"org":1098,"tags":1099,"stars":1081,"repoUrl":1082,"updatedAt":1109},"agentic-actions-auditor","audit GitHub Actions for security vulnerabilities","Audits GitHub Actions workflows for security vulnerabilities in AI agent integrations including Claude Code Action, Gemini CLI, OpenAI Codex, and GitHub AI Inference. Detects attack vectors where attacker-controlled input reaches AI agents running in CI\u002FCD pipelines, including env var intermediary patterns, direct expression injection, dangerous sandbox configurations, and wildcard user allowlists. Use when reviewing workflow files that invoke AI coding agents, auditing CI\u002FCD pipeline security for prompt injection risks, or evaluating agentic action configurations.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1100,1101,1104,1105,1108],{"name":1008,"slug":1009,"type":16},{"name":1102,"slug":1103,"type":16},"CI\u002FCD","ci-cd",{"name":961,"slug":962,"type":16},{"name":1106,"slug":1107,"type":16},"GitHub Actions","github-actions",{"name":964,"slug":965,"type":16},"2026-07-18T05:47:48.564744",{"slug":1111,"name":1111,"fn":1112,"description":1113,"org":1114,"tags":1115,"stars":1081,"repoUrl":1082,"updatedAt":1124},"algorand-vulnerability-scanner","scan Algorand smart contracts for vulnerabilities","Scans Algorand smart contracts for 11 common vulnerabilities including rekeying attacks, unchecked transaction fees, missing field validations, and access control issues. Use when auditing Algorand projects (TEAL\u002FPyTeal).",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1116,1119,1120,1121],{"name":1117,"slug":1118,"type":16},"Audit","audit",{"name":961,"slug":962,"type":16},{"name":964,"slug":965,"type":16},{"name":1122,"slug":1123,"type":16},"Smart Contracts","smart-contracts","2026-07-18T05:47:43.989063",{"slug":1126,"name":1126,"fn":1127,"description":1128,"org":1129,"tags":1130,"stars":1081,"repoUrl":1082,"updatedAt":1135},"ask-questions-if-underspecified","clarify requirements before implementation","Clarify requirements before implementing. Use when serious doubts arise.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1131,1134],{"name":1132,"slug":1133,"type":16},"Engineering","engineering",{"name":1014,"slug":1015,"type":16},"2026-07-17T06:05:33.543262",{"slug":1137,"name":1137,"fn":1138,"description":1139,"org":1140,"tags":1141,"stars":1081,"repoUrl":1082,"updatedAt":1147},"atheris","fuzz Python code with Atheris","Atheris is a coverage-guided Python fuzzer based on libFuzzer. Use for fuzzing pure Python code and Python C extensions.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1142,1145,1146],{"name":1143,"slug":1144,"type":16},"Python","python",{"name":964,"slug":965,"type":16},{"name":967,"slug":968,"type":16},"2026-07-17T06:05:14.575191",{"slug":1149,"name":1149,"fn":1150,"description":1151,"org":1152,"tags":1153,"stars":1081,"repoUrl":1082,"updatedAt":1157},"audit-augmentation","augment code graphs with audit findings","Augments Trailmark code graphs with external audit findings from SARIF static analysis results, weAudit annotation files, and version-gated Trailmark 0.4.x binary-analysis graph exports. Maps findings to graph nodes by file and line overlap, creates severity-based subgraphs, and enables cross-referencing findings with pre-analysis data (blast radius, taint, etc.). Use when projecting SARIF results onto a code graph, overlaying weAudit annotations, importing binary graph findings, cross-referencing Semgrep, CodeQL, or binary-analysis findings with call graph data, or visualizing audit findings in the context of code structure.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1154,1155,1156],{"name":1117,"slug":1118,"type":16},{"name":961,"slug":962,"type":16},{"name":964,"slug":965,"type":16},"2026-08-01T05:44:54.920542",{"slug":1159,"name":1159,"fn":1160,"description":1161,"org":1162,"tags":1163,"stars":1081,"repoUrl":1082,"updatedAt":1170},"audit-context-building","build architectural context for code analysis","Enables ultra-granular, line-by-line code analysis to build deep architectural context before vulnerability or bug finding.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1164,1167,1168,1169],{"name":1165,"slug":1166,"type":16},"Architecture","architecture",{"name":1117,"slug":1118,"type":16},{"name":961,"slug":962,"type":16},{"name":1132,"slug":1133,"type":16},"2026-07-18T05:47:40.122449",{"slug":1172,"name":1172,"fn":1173,"description":1174,"org":1175,"tags":1176,"stars":1081,"repoUrl":1082,"updatedAt":1181},"audit-prep-assistant","prepare codebases for security audits","Prepares codebases for security review using Trail of Bits' checklist. Helps set review goals, runs static analysis tools, increases test coverage, removes dead code, ensures accessibility, and generates documentation (flowcharts, user stories, inline comments).",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1177,1178,1179,1180],{"name":1117,"slug":1118,"type":16},{"name":961,"slug":962,"type":16},{"name":1132,"slug":1133,"type":16},{"name":964,"slug":965,"type":16},"2026-07-18T05:47:39.210985",{"slug":1183,"name":1183,"fn":1184,"description":1185,"org":1186,"tags":1187,"stars":1081,"repoUrl":1082,"updatedAt":1193},"burpsuite-project-parser","parse Burp Suite project files","Searches and explores Burp Suite project files (.burp) from the command line. Use when searching response headers or bodies with regex patterns, extracting security audit findings, dumping proxy history or site map data, or analyzing HTTP traffic captured in a Burp project.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1188,1189,1192],{"name":1117,"slug":1118,"type":16},{"name":1190,"slug":1191,"type":16},"CLI","cli",{"name":964,"slug":965,"type":16},"2026-07-17T06:05:33.198077",{"slug":1195,"name":1195,"fn":1196,"description":1197,"org":1198,"tags":1199,"stars":1081,"repoUrl":1082,"updatedAt":1204},"c-review","audit C and C++ code","Performs comprehensive C\u002FC++ security review for memory corruption, integer overflows, race conditions, and platform-specific vulnerabilities. Use when auditing native C\u002FC++ applications, reviewing daemons or services for memory safety, or hunting integer overflow \u002F use-after-free \u002F race conditions in userspace code.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1200,1201,1202,1203],{"name":1117,"slug":1118,"type":16},{"name":1076,"slug":1077,"type":16},{"name":961,"slug":962,"type":16},{"name":964,"slug":965,"type":16},"2026-07-17T06:05:11.333374",{"slug":1206,"name":1206,"fn":1207,"description":1208,"org":1209,"tags":1210,"stars":1081,"repoUrl":1082,"updatedAt":1215},"cairo-vulnerability-scanner","scan Cairo and StarkNet contracts for vulnerabilities","Scans Cairo\u002FStarkNet smart contracts for 6 critical vulnerabilities including felt252 arithmetic overflow, L1-L2 messaging issues, address conversion problems, and signature replay. Use when auditing StarkNet projects.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1211,1212,1213,1214],{"name":1117,"slug":1118,"type":16},{"name":961,"slug":962,"type":16},{"name":964,"slug":965,"type":16},{"name":1122,"slug":1123,"type":16},"2026-07-18T05:47:42.84568",111]