[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M718VJRMSQJ14CXAR99K3AJC":3,"skill-same:01M718VJRMSQJ14CXAR99K3AJC":70},{"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":23,"publishedAt":44,"linkOnly":45,"markdownContent":46,"markdownBody":47,"frontMatter":48,"refName":49,"commitSha":50,"skillDirectory":51,"forksCount":14,"repositoryHtmlUrl":52,"repositoryHomepageUrl":6,"ownerHtmlUrl":53,"ownerBlogUrl":6,"ownerDescription":54,"updatedAt":55,"install":56,"resources":69},"01M718VJRMSQJ14CXAR99K3AJC","typesafe-ai",null,"Build AI-powered software with TypeSafe: small units of AI intelligence you can use like programming primitives. Its System One models, including Jev, turn natural language and application state into typed judgments and probabilities that code can combine. Use when a feature needs programmable common sense, when brainstorming what AI could make possible in an app, or when an LLM prompt-and-parse step could become a structured decision. Applications include routing, ranking, extraction, verification, and interactive experiences; these are starting points, not the limits. Read live docs and cookbooks to find useful patterns and discover new combinations.","SigurdHall\u002Fskills","SigurdHall","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F218090219?v=4","01M718QDXXHGFJ3B5RHAHB6GX3","Sigurd Hall","01M718QDYY3AHN7NRJKBED94ZA",0,"MIT","https:\u002F\u002Fgithub.com\u002FSigurdHall\u002Fskills\u002Fblob\u002Fmaster\u002Fskills\u002Fcoding\u002Ftypesafe-ai\u002FSKILL.md",[18],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000030","开发工程师","PROFESSION","适用职业",[24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43],"typesafe ai","common","primitives","patterns","live","intelligence","software","read","useful","limits","verification","code","you","when","new","ranking","decision","parse","discover","app","2026-11-10 10:12:30",false,"---\nname: typesafe-ai\nlicense: MIT\ndescription: >\n  Build AI-powered software with TypeSafe: small units of AI intelligence you\n  can use like programming primitives. Its System One models, including Jev,\n  turn natural language and application state into typed judgments and\n  probabilities that code can combine. Use when a feature needs programmable\n  common sense, when brainstorming what AI could make possible in an app, or\n  when an LLM prompt-and-parse step could become a structured decision.\n  Applications include routing, ranking, extraction, verification, and\n  interactive experiences; these are starting points, not the limits.\n  Read live docs and cookbooks to find useful patterns and discover new combinations.\n---\n\n# Build with TypeSafe\n\nTypeSafe makes units of AI intelligence usable like programming primitives: small\njudgments you can compose into larger capabilities. Its **System One models** return\nfast, focused judgments that software can consume directly. **Jev** is TypeSafe's\nflagship and first System One model. It understands natural language and returns\ntyped answers and probabilities rather\nthan generating text or reasoning explanations. Code owns the workflow; the model\nsupplies programmable common sense where ordinary code needs semantic understanding.\n\n## Read the live docs\n\n**The live TypeSafe docs are the source of truth. Read them as part of the task.**\nThis skill gives direction; the docs carry current concepts, prompting guidance,\nAPI contracts, SDK usage, models, limits, and worked examples.\n\n- Start with the [documentation index](https:\u002F\u002Fdocs.typesafe.ai\u002Fllms.txt) to discover\n  relevant pages and cookbooks. Use targeted reads rather than loading the entire site.\n- Mintlify serves Markdown by appending `.md` to a page path, for example\n  [how to build with TypeSafe](https:\u002F\u002Fdocs.typesafe.ai\u002Fconcepts\u002Fhow-to-build-with-system-one.md).\n  Follow links from the index; convert extensionless documentation page links to\n  `.md` when useful. Resolve relative links against `https:\u002F\u002Fdocs.typesafe.ai`.\n- Before writing an integration, read the current API or chosen SDK page and the\n  question guidance relevant to the design. For a new workflow, also inspect the\n  closest cookbook: it often shows a better decomposition than a generic classifier.\n- If the index is unavailable, use the direct links below or the site's navigation.\n  If Markdown fetching fails, try the normal page. If live access is unavailable,\n  use available local docs or installed SDK types, state that limitation, and avoid\n  inventing version-dependent details.\n\n| Task | Start here; follow the relevant details |\n| --- | --- |\n| Understand the programming model | [System One](https:\u002F\u002Fdocs.typesafe.ai\u002Fconcepts\u002Fsystem-one.md), [building guide](https:\u002F\u002Fdocs.typesafe.ai\u002Fconcepts\u002Fhow-to-build-with-system-one.md) |\n| Explore what to build | [Use-case map](https:\u002F\u002Fdocs.typesafe.ai\u002Fconcepts\u002Fuse-case-map.md), then relevant cookbooks from the index |\n| Prepare inputs and questions | [State](https:\u002F\u002Fdocs.typesafe.ai\u002Fconcepts\u002Fstate.md), [primitives](https:\u002F\u002Fdocs.typesafe.ai\u002Fprimitives.md), then the chosen primitive's page |\n| Decide how to handle uncertainty | [Confidence](https:\u002F\u002Fdocs.typesafe.ai\u002Fconfidence.md) |\n| Write API code | [HTTP API](https:\u002F\u002Fdocs.typesafe.ai\u002Fapi.md), [Python SDK](https:\u002F\u002Fdocs.typesafe.ai\u002Fsdk\u002Fpython.md), or [JavaScript SDK](https:\u002F\u002Fdocs.typesafe.ai\u002Fsdk\u002Fjavascript.md) |\n| Update an older integration | [Migration guide](https:\u002F\u002Fdocs.typesafe.ai\u002Fmigrating-to-v1.md) and the installed SDK's current reference |\n\n## Find the useful shape\n\nStart from the behavior the user wants: what will the application show, select,\nchange, or hand off? Work backward to the judgments it needs. Keep known rules,\ncalculations, exact lookups, and execution in code. Preserve the user's chosen stack\nand scope; add TypeSafe where semantic understanding helps.\n\nWhen brainstorming or choosing an architecture, consider more than classification.\nThe patterns below are starting points: combine primitives around the user's goal,\nincluding ideas that do not fit an established recipe.\n\n- **Route and fill known arguments.** A request can select a handler and its typed\n  parameters. Ask useful branch-specific questions up front and consume only the\n  relevant answers. Explore [function calling](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Ffunction_calling.md)\n  and [speculative fan-out](https:\u002F\u002Fdocs.typesafe.ai\u002Fpatterns\u002Ffan-out.md).\n- **Select instead of generate.** Find candidate values or source spans in code,\n  use a judgment to select the intended one, then copy or normalize it. Code can\n  also assemble source text into a formatted document or reading guide. Explore\n  [value extraction](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Fpre_parsed_value_extraction_cookbook.md)\n  and [structure recovery](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Fautoformat.md).\n- **Find and judge evidence.** Retrieve candidates, compare their relevance to a\n  query, and select useful context. Explore [reranking](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Frerank_typesafe.md)\n  and [hierarchical classification](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Fhierarchical_classification.md).\n- **Turn judgments into reusable data.** Score dimensions once, then let code or\n  user controls change weights, thresholds, rankings, and views. With labeled\n  outcomes, those signals can become classical ML features. Explore\n  [composite scoring](https:\u002F\u002Fdocs.typesafe.ai\u002Fpatterns\u002Fcomposite-scoring.md) and\n  [feature discovery](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Fautoresearch_feature_discovery.md).\n- **Verify and escalate.** Check specific claims or fields against their evidence;\n  send uncertain or failing cases to a person or reasoning model. Explore\n  [citation checks](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Fcitation_check.md) and\n  [extraction cascades](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Fsde_cascade.md).\n- **Respond to changing state.** Code can retain goals and observations while fresh\n  judgments guide the next bounded step. Keep inferred state distinct from observed\n  facts, and check freshness before applying a result to a changed situation.\n\nFor open-ended requests, offer the few directions that best serve the user's goal\nand recommend a starting point. For a concrete request, choose the relevant pattern\nand build; a brainstorm is not a mandatory detour.\n\n## Design the judgments\n\nChoose by what the answer means, then read the relevant primitive page:\n\n| Need | Primitive | Important distinction |\n| --- | --- | --- |\n| One of a defined set | [Choice](https:\u002F\u002Fdocs.typesafe.ai\u002Fprimitives\u002Fchoice.md) | Picks one option; its distribution compares competing options |\n| Whether a condition holds | [Noul](https:\u002F\u002Fdocs.typesafe.ai\u002Fprimitives\u002Fnoul.md) | Probability of yes; no separate confidence; use one per label when several may apply |\n| Degree along a described dimension | [Score](https:\u002F\u002Fdocs.typesafe.ai\u002Fprimitives\u002Fscore.md) | Probability-weighted position on ordered levels; use comparable per-item Scores for graded ranking |\n\nGive each question enough relevant **state** to answer: source text, identities,\nrelationships, policies, and current facts. Prefer named JSON fields when context\nhas several parts. Put the judgment in **instructions** and define its possible\nanswers in **criteria**. Question IDs are for code and are not sent to the model;\ninclude complete meaning in the question. Reference nested state with backticked\npaths such as `ticket.messages[0].text`.\n\nAsk one narrow, coherent judgment per question. Split independently useful dimensions,\nwithout destroying the relationship being judged. A bounded action selection or\ncontextual interpretation is valid; atomic does not mean literal fact extraction\nor a one-sentence limit. Strings work for simple questions. Use structured objects\nor arrays when definitions, contrasts, exclusions, or examples clarify instructions\nor criteria. Score levels must describe concrete situations and stand on their own.\n\nKeep the needed answers available. Include a no-match outcome when nothing may fit;\nuse a separate presence judgment when it is independently useful. For source-value\nselection, check candidate coverage: the model cannot choose an omitted value.\n\n## Compose and verify\n\n**Ask independent questions over the same state together**, including useful\nspeculative questions. They run in parallel and cannot see one another's answers.\nState each speculative premise explicitly; code consumes the applicable answers.\nA second request is warranted when an earlier answer is needed to fetch evidence,\nconstruct new state, or determine the next options. Extra questions still use tokens;\nmeasure actual request budgets, cost, and end-to-end latency.\n\nUse probabilities and confidence to guide behavior, with thresholds evaluated on\nthe user's data and consequences. Choice\u002FScore confidence summarizes distribution\nconcentration, not overall workflow correctness or permission to act. A Noul near\n0.5 means similar probability for yes and no, not medium intensity. Several\nacceptable alternatives can also spread probability; low confidence need not\ninvalidate a harmless preference choice. Ignore uncertainty on unused branches.\n\nKeep policy explicit and raw judgments reusable. Weighted scores suit compensating\npreferences; an “any serious violation” rule needs separate conditions. Changing a\nweight or display filter need not rerun inference when evidence and question meanings\nare unchanged. Typed output guarantees the interface, not truth. System One models\nare trained for calibrated decisions; validate their performance in the target domain.\n\nTest representative cases and the resulting application behavior. For failures,\ninspect the exact state, questions, candidates, answers, composition, and observed\noutcome. Separate missing evidence, model errors, code errors, and service failures.\nTreat cookbook thresholds and demo results as examples to evaluate, not universal\nrules or permanent model limitations. Keep API credentials server-side in web apps.\n","# Build with TypeSafe\n\nTypeSafe makes units of AI intelligence usable like programming primitives: small\njudgments you can compose into larger capabilities. Its **System One models** return\nfast, focused judgments that software can consume directly. **Jev** is TypeSafe's\nflagship and first System One model. It understands natural language and returns\ntyped answers and probabilities rather\nthan generating text or reasoning explanations. Code owns the workflow; the model\nsupplies programmable common sense where ordinary code needs semantic understanding.\n\n## Read the live docs\n\n**The live TypeSafe docs are the source of truth. Read them as part of the task.**\nThis skill gives direction; the docs carry current concepts, prompting guidance,\nAPI contracts, SDK usage, models, limits, and worked examples.\n\n- Start with the [documentation index](https:\u002F\u002Fdocs.typesafe.ai\u002Fllms.txt) to discover\n  relevant pages and cookbooks. Use targeted reads rather than loading the entire site.\n- Mintlify serves Markdown by appending `.md` to a page path, for example\n  [how to build with TypeSafe](https:\u002F\u002Fdocs.typesafe.ai\u002Fconcepts\u002Fhow-to-build-with-system-one.md).\n  Follow links from the index; convert extensionless documentation page links to\n  `.md` when useful. Resolve relative links against `https:\u002F\u002Fdocs.typesafe.ai`.\n- Before writing an integration, read the current API or chosen SDK page and the\n  question guidance relevant to the design. For a new workflow, also inspect the\n  closest cookbook: it often shows a better decomposition than a generic classifier.\n- If the index is unavailable, use the direct links below or the site's navigation.\n  If Markdown fetching fails, try the normal page. If live access is unavailable,\n  use available local docs or installed SDK types, state that limitation, and avoid\n  inventing version-dependent details.\n\n| Task | Start here; follow the relevant details |\n| --- | --- |\n| Understand the programming model | [System One](https:\u002F\u002Fdocs.typesafe.ai\u002Fconcepts\u002Fsystem-one.md), [building guide](https:\u002F\u002Fdocs.typesafe.ai\u002Fconcepts\u002Fhow-to-build-with-system-one.md) |\n| Explore what to build | [Use-case map](https:\u002F\u002Fdocs.typesafe.ai\u002Fconcepts\u002Fuse-case-map.md), then relevant cookbooks from the index |\n| Prepare inputs and questions | [State](https:\u002F\u002Fdocs.typesafe.ai\u002Fconcepts\u002Fstate.md), [primitives](https:\u002F\u002Fdocs.typesafe.ai\u002Fprimitives.md), then the chosen primitive's page |\n| Decide how to handle uncertainty | [Confidence](https:\u002F\u002Fdocs.typesafe.ai\u002Fconfidence.md) |\n| Write API code | [HTTP API](https:\u002F\u002Fdocs.typesafe.ai\u002Fapi.md), [Python SDK](https:\u002F\u002Fdocs.typesafe.ai\u002Fsdk\u002Fpython.md), or [JavaScript SDK](https:\u002F\u002Fdocs.typesafe.ai\u002Fsdk\u002Fjavascript.md) |\n| Update an older integration | [Migration guide](https:\u002F\u002Fdocs.typesafe.ai\u002Fmigrating-to-v1.md) and the installed SDK's current reference |\n\n## Find the useful shape\n\nStart from the behavior the user wants: what will the application show, select,\nchange, or hand off? Work backward to the judgments it needs. Keep known rules,\ncalculations, exact lookups, and execution in code. Preserve the user's chosen stack\nand scope; add TypeSafe where semantic understanding helps.\n\nWhen brainstorming or choosing an architecture, consider more than classification.\nThe patterns below are starting points: combine primitives around the user's goal,\nincluding ideas that do not fit an established recipe.\n\n- **Route and fill known arguments.** A request can select a handler and its typed\n  parameters. Ask useful branch-specific questions up front and consume only the\n  relevant answers. Explore [function calling](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Ffunction_calling.md)\n  and [speculative fan-out](https:\u002F\u002Fdocs.typesafe.ai\u002Fpatterns\u002Ffan-out.md).\n- **Select instead of generate.** Find candidate values or source spans in code,\n  use a judgment to select the intended one, then copy or normalize it. Code can\n  also assemble source text into a formatted document or reading guide. Explore\n  [value extraction](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Fpre_parsed_value_extraction_cookbook.md)\n  and [structure recovery](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Fautoformat.md).\n- **Find and judge evidence.** Retrieve candidates, compare their relevance to a\n  query, and select useful context. Explore [reranking](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Frerank_typesafe.md)\n  and [hierarchical classification](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Fhierarchical_classification.md).\n- **Turn judgments into reusable data.** Score dimensions once, then let code or\n  user controls change weights, thresholds, rankings, and views. With labeled\n  outcomes, those signals can become classical ML features. Explore\n  [composite scoring](https:\u002F\u002Fdocs.typesafe.ai\u002Fpatterns\u002Fcomposite-scoring.md) and\n  [feature discovery](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Fautoresearch_feature_discovery.md).\n- **Verify and escalate.** Check specific claims or fields against their evidence;\n  send uncertain or failing cases to a person or reasoning model. Explore\n  [citation checks](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Fcitation_check.md) and\n  [extraction cascades](https:\u002F\u002Fdocs.typesafe.ai\u002Fcookbooks\u002Fsde_cascade.md).\n- **Respond to changing state.** Code can retain goals and observations while fresh\n  judgments guide the next bounded step. Keep inferred state distinct from observed\n  facts, and check freshness before applying a result to a changed situation.\n\nFor open-ended requests, offer the few directions that best serve the user's goal\nand recommend a starting point. For a concrete request, choose the relevant pattern\nand build; a brainstorm is not a mandatory detour.\n\n## Design the judgments\n\nChoose by what the answer means, then read the relevant primitive page:\n\n| Need | Primitive | Important distinction |\n| --- | --- | --- |\n| One of a defined set | [Choice](https:\u002F\u002Fdocs.typesafe.ai\u002Fprimitives\u002Fchoice.md) | Picks one option; its distribution compares competing options |\n| Whether a condition holds | [Noul](https:\u002F\u002Fdocs.typesafe.ai\u002Fprimitives\u002Fnoul.md) | Probability of yes; no separate confidence; use one per label when several may apply |\n| Degree along a described dimension | [Score](https:\u002F\u002Fdocs.typesafe.ai\u002Fprimitives\u002Fscore.md) | Probability-weighted position on ordered levels; use comparable per-item Scores for graded ranking |\n\nGive each question enough relevant **state** to answer: source text, identities,\nrelationships, policies, and current facts. Prefer named JSON fields when context\nhas several parts. Put the judgment in **instructions** and define its possible\nanswers in **criteria**. Question IDs are for code and are not sent to the model;\ninclude complete meaning in the question. Reference nested state with backticked\npaths such as `ticket.messages[0].text`.\n\nAsk one narrow, coherent judgment per question. Split independently useful dimensions,\nwithout destroying the relationship being judged. A bounded action selection or\ncontextual interpretation is valid; atomic does not mean literal fact extraction\nor a one-sentence limit. Strings work for simple questions. Use structured objects\nor arrays when definitions, contrasts, exclusions, or examples clarify instructions\nor criteria. Score levels must describe concrete situations and stand on their own.\n\nKeep the needed answers available. Include a no-match outcome when nothing may fit;\nuse a separate presence judgment when it is independently useful. For source-value\nselection, check candidate coverage: the model cannot choose an omitted value.\n\n## Compose and verify\n\n**Ask independent questions over the same state together**, including useful\nspeculative questions. They run in parallel and cannot see one another's answers.\nState each speculative premise explicitly; code consumes the applicable answers.\nA second request is warranted when an earlier answer is needed to fetch evidence,\nconstruct new state, or determine the next options. Extra questions still use tokens;\nmeasure actual request budgets, cost, and end-to-end latency.\n\nUse probabilities and confidence to guide behavior, with thresholds evaluated on\nthe user's data and consequences. Choice\u002FScore confidence summarizes distribution\nconcentration, not overall workflow correctness or permission to act. A Noul near\n0.5 means similar probability for yes and no, not medium intensity. Several\nacceptable alternatives can also spread probability; low confidence need not\ninvalidate a harmless preference choice. Ignore uncertainty on unused branches.\n\nKeep policy explicit and raw judgments reusable. Weighted scores suit compensating\npreferences; an “any serious violation” rule needs separate conditions. Changing a\nweight or display filter need not rerun inference when evidence and question meanings\nare unchanged. Typed output guarantees the interface, not truth. System One models\nare trained for calibrated decisions; validate their performance in the target domain.\n\nTest representative cases and the resulting application behavior. For failures,\ninspect the exact state, questions, candidates, answers, composition, and observed\noutcome. Separate missing evidence, model errors, code errors, and service failures.\nTreat cookbook thresholds and demo results as examples to evaluate, not universal\nrules or permanent model limitations. Keep API credentials server-side in web apps.\n",{"name":5,"license":15,"description":7},"master","f47f04a75f2ff283ea6b09c9413cee66af1a68df","skills\u002Fcoding\u002Ftypesafe-ai","https:\u002F\u002Fgithub.com\u002FSigurdHall\u002Fskills","https:\u002F\u002Fgithub.com\u002FSigurdHall","Finance and BI at a Norwegian university. Data, reporting, and making numbers useful.","2026-11-03 20:01:33",{"source":8,"skillName":5,"commands":57,"prompt":67,"zipAvailable":68},[58,61,64],{"tool":59,"command":60},"npx","npx skills add SigurdHall\u002Fskills --skill typesafe-ai",{"tool":62,"command":63},"bunx","bunx skills add SigurdHall\u002Fskills --skill typesafe-ai",{"tool":65,"command":66},"pnpm","pnpm dlx skills add SigurdHall\u002Fskills --skill typesafe-ai","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FSigurdHall\u002Fskills\u002Fblob\u002Fmaster\u002Fskills\u002Fcoding\u002Ftypesafe-ai\u002FSKILL.md",true,[],[]]