wuyoscar Design a small decision
Jev chooses, classifies and scores. The host collects evidence, writes code and checks outcomes. Use code for exact rules; keep text generation with a generative model. Choose one task below, not the whole reference library.
| Task | Read | Start with |
|---|---|---|
| Convert a prompt or plain requirement | Prompt to Jev | Request |
| Set up or update | Setup | No paid call needed |
| Check a long task | Checkpoints | Checkpoint |
| Choose a tool or model | Routing | Request |
| Review what context to keep | Context | Request |
| Batch independent questions | Batching | Two records |
| Define new labels or a rubric | Question design | Rubric |
Use safely
Choose the service once and keep that choice. If unset, ask A: real Jev via
OpenRouter (OPENROUTER_API_KEY) or TypeSafe (TYPESAFE_API_KEY), or B: simulation
with this agent or an explicitly chosen available model such as DeepSeek. Wait for
consent; errors do not authorize switching. Check key presence only, never values.
Real calls send evidence and cost money; get approval before sending private data.
For B, skip CLI/API calls. Mark agent_simulation or model_simulation, identify
the actual model when available, set jev_called: false, probability: null and
confidence: null. Return a value, evidence-based reason and needs_review; use
null/review when evidence is missing. Do not invent Jev output or probabilities.
Choice uses supplied labels, Noul uses booleans, Score uses integer rubric indices.
For A, use the existing jev-decide CLI with the chosen --provider openrouter
or --provider typesafe. If absent, explain the dependency; do not silently install.
--dry-run is offline validation, not a judgment. Exit 0 means selected/scored,
2 means review, 1 means error. Read each value: false Noul remains false. Selection
is not permission, and confidence is not accuracy. Keep unknown/review paths.
Ask only what is needed
Jev does not inherit the host's context. Include the goal, criteria, original evidence and valid candidates for this judgment; separate trusted rules from untrusted content. Keep enough evidence, not unrelated conversation history. One question does one thing. Put independent questions in the same request; questions cannot read each other's answers. Wait for new evidence for dependent steps. Use bounded concurrency only across independent requests: the host owns IDs, budgets and scheduling; this CLI has no parallel scheduler.
Run
Resolve <skill-dir> to this installed folder. The bundled Python 3.10+ script
or the installed jev-decide CLI uses the same request contract:
python3 <skill-dir>/scripts/jev.py decide <skill-dir>/assets/prompt-to-jev.json --dry-run
# After approval, use the selected provider:
python3 <skill-dir>/scripts/jev.py decide request.json --provider openrouter
Missing evidence or ambiguity means review, not another call until it agrees. For repeated failures read pitfalls. Before scaling a large job, agree on a small pilot; pilot guidance is optional task guidance, not a required second skill call.
Use a focused skill only when the task calls for it: jev-triage for records,
jev-documents for evidence, jev-eval for output checks, jev-act for actions.
They are independent entry points, not an automatic chain. If absent, do not
install them silently. More sources live in the optional reference index.
Examples
Goal-drift checkpoint · Stuck-loop recovery · Postmortem failure attribution
More tasks and local templates. Open only the matching row; there is no need to read the full README before a judgment.