decide
FeaturedUse when the user wants to try BlockRun's free typed-judgment endpoint (POST api.blockrun.ai/v1/decide, served by OpenJev) on their own data from Claude Code — yes/no, labelled choice, or scored-rung questions over a text or JSON state, up to 64 per call. Not an MCP tool: call it with curl from the shell. Covers the request shape, what the confidence number does and does not mean, and why OpenJev is not Jev.
Install
Quality Score: 93/100
Skill Content
Details
- Author
- BlockRunAI
- Repository
- BlockRunAI/blockrun-mcp
- Created
- 8 months ago
- Last Updated
- yesterday
- Language
- TypeScript
- License
- MIT
Integrates with
Bundled in these plugins
Similar Skills
Semantically similar based on skill content — not just same category
decision-model-batch
Run massive closed questions (classify, filter, tag, route, score, yes/no, link or not) over hundreds or thousands of items at near-zero cost with a typed decision model instead of a text LLM - Jev by TypeSafe through OpenRouter (`typesafe/jev-1.13`, endpoint POST https://openrouter.ai/api/alpha/decisions, not chat/completions). Covers the exact call format, the three question types (noul, choice, score), how to write questions, stacking dozens of questions per call, calibrating on a sample then deciding in code with explicit thresholds, caching raw probabilities, cost and throughput, and the pitfalls. Use it whenever a task is "the same closed question N times", when the user mentions Jev, TypeSafe or a decision model, or complains about what an LLM costs on a bulk job, even if they just say "classify all these" or "filter this list".
jev
Use the Jev decision model (TypeSafe System One, via OpenRouter) for bounded decisions inside coding workflows — routing, triage, classification, gating, rubric grading, ranking, and reducing large data before it reaches the main model. Use when a decision has a fixed set of possible answers, when data is too large or too noisy to put in context, when the same judgement must be made many times, or when an irreversible action needs a cheap safety check. Jev emits no text: never use it for prose, code generation, summaries or reasoning. Triggers: "classify", "categorize", "which of these", "route", "triage", "gate", "should we", "rank", "prioritize", "grade", "too many logs", "reduce the data", "save tokens", "batch decisions", "is it safe to". Advisory, never an authorization boundary: Jev can be wrong, manipulated or overconfident, so do not map a returned label straight to an irreversible or destructive action without your own deterministic check.
jev
Use TypeSafe's Jev (System One) model for fast, cheap, calibrated typed decisions — a yes/no probability (noul), a pick from fixed options (choice), or a position on a rubric (score). Two modes: (1) inside a Claude Code session, call Jev through the bundled script to bulk-judge many items at once (triage logcat lines, crash groups, lint warnings, PR diffs, string resources, file relevance, review comments); (2) when building software, design and write correct Jev integrations (routing, classification, guardrails, scoring, gating, reranking) in Kotlin/Android, Swift/iOS, backends, Python or JS. Use this skill whenever the user mentions Jev, TypeSafe, System One, noul, typed or calibrated decisions, "fuzzy if", confidence-gated routing, or wants to classify, score, rank, flag, calibrate, or triage a batch of text items — even if they don't name Jev.