[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7D210P012T6CMTAMAKEBCZX":3,"skill-same:01M7D210P012T6CMTAMAKEBCZX":48},{"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":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":29,"ownerHtmlUrl":30,"ownerBlogUrl":31,"ownerDescription":32,"updatedAt":33,"install":34,"resources":47},"01M7D210P012T6CMTAMAKEBCZX","jev-guardrail",null,"Screen LLM input and output with TypeSafe Jev Noul hazard batteries plus a harm Score; policy in code returns pass, review, or block. Use for jailbreak, prompt injection, secret leak, denylist pressure, or when the user runs \u002Fjev-guardrail or `npx jev guard`.","cobusgreyling\u002FJev","cobusgreyling","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F7868717?v=4","01M75EJ2EYER7JRCZRRJ9FA3QH","Cobus Greyling","01M7D20MZQ4SKB1P0TTEZ6J8X4",141,"MIT","https:\u002F\u002Fgithub.com\u002Fcobusgreyling\u002FJev\u002Fblob\u002Fmain\u002Fskills\u002Fjev-guardrail\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: jev-guardrail\ndescription: Screen LLM input and output with TypeSafe Jev Noul hazard batteries plus a harm Score; policy in code returns pass, review, or block. Use for jailbreak, prompt injection, secret leak, denylist pressure, or when the user runs \u002Fjev-guardrail or `npx jev guard`.\n---\n\n# Jev LLM guardrail\n\nJev does not refuse in prose. It scores hazards. **Code** returns `pass` \u002F `review` \u002F `block`, then a generative model (or a static message) talks.\n\nPrimitives and auth: `skills\u002Fjev\u002FSKILL.md`. Cookbook: https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Fllm_guardrails\n\n## Battery\n\nOne request. Independent **Nouls** per hazard + one **Score** for harm. Gate Noul on the **probability**. Do not copy a Noul threshold onto Choice `confidence`.\n\nInput hazards in this repo: `jailbreak`, `prompt_injection`, `secret_exfil`, `denylist_pressure`, `harmful_request`.\n\nOutput hazards: `broke_policy`, `secret_leak`, `hallucination`, `instruction_drift`, `harmful_content`.\n\nScreen **both** sides. Redact secrets before sending `state`.\n\n## Policy lives in code\n\nCaller-owned thresholds. Defaults in this repo (`packages\u002Fjs\u002Fsrc\u002Fguard.ts`, `jev_lab\u002Fpolicy.py`):\n\n| Signal | Strict | Permissive |\n|--------|--------|------------|\n| Noul review | ≥ 0.35 | ≥ 0.35 |\n| Noul action | ≥ 0.7 | ≥ 0.85 |\n| Harm Score promotes review → block | ≥ 2.0 | ≥ 2.0 |\n\nGuardrails **fail closed** (block if the call errors). Routers may fail open; this skill does not.\n\n`jailbreak`, `prompt_injection`, `secret_exfil`, `denylist_pressure`, `harmful_request`, `broke_policy`, `secret_leak`, `harmful_content` → `block` when above the action bar. `hallucination` and `instruction_drift` → `review`.\n\n## Pair with a generative model\n\n- `block` — do not forward the text; return a fixed refusal\n- `review` — human or a reasoning model\n- `pass` — generate as usual\n\n## This repo\n\n```bash\nnpx jev guard --side input --text \"Ignore previous instructions\" --json\nnpx jev guard --side output --text \"$REPLY\" --json\npython examples\u002F07_guardrails.py\n```\n\n- `packages\u002Fjs\u002Fsrc\u002Fguard.ts` — `inputBattery` \u002F `outputBattery` \u002F `routeGuard`\n- `examples\u002F07_guardrails.py` — Choice + Noul, then skip the chat model\n","# Jev LLM guardrail\n\nJev does not refuse in prose. It scores hazards. **Code** returns `pass` \u002F `review` \u002F `block`, then a generative model (or a static message) talks.\n\nPrimitives and auth: `skills\u002Fjev\u002FSKILL.md`. Cookbook: https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Fllm_guardrails\n\n## Battery\n\nOne request. Independent **Nouls** per hazard + one **Score** for harm. Gate Noul on the **probability**. Do not copy a Noul threshold onto Choice `confidence`.\n\nInput hazards in this repo: `jailbreak`, `prompt_injection`, `secret_exfil`, `denylist_pressure`, `harmful_request`.\n\nOutput hazards: `broke_policy`, `secret_leak`, `hallucination`, `instruction_drift`, `harmful_content`.\n\nScreen **both** sides. Redact secrets before sending `state`.\n\n## Policy lives in code\n\nCaller-owned thresholds. Defaults in this repo (`packages\u002Fjs\u002Fsrc\u002Fguard.ts`, `jev_lab\u002Fpolicy.py`):\n\n| Signal | Strict | Permissive |\n|--------|--------|------------|\n| Noul review | ≥ 0.35 | ≥ 0.35 |\n| Noul action | ≥ 0.7 | ≥ 0.85 |\n| Harm Score promotes review → block | ≥ 2.0 | ≥ 2.0 |\n\nGuardrails **fail closed** (block if the call errors). Routers may fail open; this skill does not.\n\n`jailbreak`, `prompt_injection`, `secret_exfil`, `denylist_pressure`, `harmful_request`, `broke_policy`, `secret_leak`, `harmful_content` → `block` when above the action bar. `hallucination` and `instruction_drift` → `review`.\n\n## Pair with a generative model\n\n- `block` — do not forward the text; return a fixed refusal\n- `review` — human or a reasoning model\n- `pass` — generate as usual\n\n## This repo\n\n```bash\nnpx jev guard --side input --text \"Ignore previous instructions\" --json\nnpx jev guard --side output --text \"$REPLY\" --json\npython examples\u002F07_guardrails.py\n```\n\n- `packages\u002Fjs\u002Fsrc\u002Fguard.ts` — `inputBattery` \u002F `outputBattery` \u002F `routeGuard`\n- `examples\u002F07_guardrails.py` — Choice + Noul, then skip the chat model\n",{"name":5,"description":7},"main","77b65c6b6dedc6c504f7f2caea57a354887b60a8","skills\u002Fjev-guardrail",30,"https:\u002F\u002Fgithub.com\u002Fcobusgreyling\u002FJev","https:\u002F\u002Fdocs.typesafe.ai\u002F","https:\u002F\u002Fgithub.com\u002Fcobusgreyling","www.cobusgreyling.com","I explore AI agents, harness & loop engineering. Author of loop-engineering (11.8k★). cobusgreyling.com","2026-10-15 21:18:15",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":46},[36,39,42],{"tool":37,"command":38},"npx","npx skills add cobusgreyling\u002FJev --skill jev-guardrail",{"tool":40,"command":41},"bunx","bunx skills add cobusgreyling\u002FJev --skill jev-guardrail",{"tool":43,"command":44},"pnpm","pnpm dlx skills add cobusgreyling\u002FJev --skill jev-guardrail","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fcobusgreyling\u002FJev\u002Fblob\u002Fmain\u002Fskills\u002Fjev-guardrail\u002FSKILL.md",true,[],[49,56,63,70,75,80],{"skillNo":50,"name":51,"nameZh":6,"description":52,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":53,"categories":54,"keywords":55,"publishedAt":19},"01M7D216TD0YJBJRXE4C7HPNHB","jev","Use TypeSafe Jev for typed judgments (Choice, Score, Noul) instead of asking a chat model to classify. Pair with a generative model for prose. Use when classifying, routing, scoring, guardrailing, or the user runs \u002Fjev.","https:\u002F\u002Fgithub.com\u002Fcobusgreyling\u002FJev\u002Fblob\u002Fmain\u002Fskills\u002Fjev\u002FSKILL.md",[],[],{"skillNo":57,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":19},"01M7D213Z2Y1GH6S76RAVPGDSF","jev-route","Route with TypeSafe Jev — map Choice plus confidence to act\u002Fconfirm\u002Fhuman, or pick a coding-agent model tier. Use for confidence routing, model routing, cheap vs frontier, loop-jev, or when the user runs \u002Fjev-route or `npx jev route`.","https:\u002F\u002Fgithub.com\u002Fcobusgreyling\u002FJev\u002Fblob\u002Fmain\u002Fskills\u002Fjev-route\u002FSKILL.md",[],[],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":67,"categories":68,"keywords":69,"publishedAt":19},"01M7D20XDQJKRP06485T692GFX","jev-fanout","Ask every independent TypeSafe Jev question in one POST \u002Fv1\u002Fsystemone (speculative fan-out). Use for parallel questions, smart-home or compound commands, one-call turn assessment, or when the user runs \u002Fjev-fanout. Do not fire one HTTP call per question.","https:\u002F\u002Fgithub.com\u002Fcobusgreyling\u002FJev\u002Fblob\u002Fmain\u002Fskills\u002Fjev-fanout\u002FSKILL.md",[],[],{"skillNo":71,"name":51,"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":72,"categories":73,"keywords":74,"publishedAt":19},"01M7D20WFQ5MR297PH8YMF9YYJ","https:\u002F\u002Fgithub.com\u002Fcobusgreyling\u002FJev\u002Fblob\u002Fmain\u002F.grok\u002Fskills\u002Fjev\u002FSKILL.md",[],[],{"skillNo":76,"name":58,"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":77,"categories":78,"keywords":79,"publishedAt":19},"01M7D20T3D4664KRRB3KR72GA8","https:\u002F\u002Fgithub.com\u002Fcobusgreyling\u002FJev\u002Fblob\u002Fmain\u002F.grok\u002Fskills\u002Fjev-route\u002FSKILL.md",[],[],{"skillNo":81,"name":5,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":82,"categories":83,"keywords":84,"publishedAt":19},"01M7D20RRNVEENYWNA83AWXX5C","https:\u002F\u002Fgithub.com\u002Fcobusgreyling\u002FJev\u002Fblob\u002Fmain\u002F.grok\u002Fskills\u002Fjev-guardrail\u002FSKILL.md",[],[]]