jev-predict-skill
Agent skill that predicts which closed-set outcome another skill would reach, by asking Jev instead of running that skill.
This repository is mostly documentation or write-up rather than runnable code.
Overview
jev-predict-skill is a single SKILL.md host-agent skill. Point it at another skill and it reports the discrete outcome that skill would pick right now, such as SHIP or NEEDS_WORK, with a confidence and a probability per outcome, by asking Jev rather than executing the target skill.
It resolves the target, gathers compact read-only evidence the target's next decision depends on, and first asks a Noul whether the skill reduces to a closed set of outcomes; below 0.65 it reports that it cannot break the skill down. Otherwise it extracts the outcome labels from the target skill itself, adds none_of_the_above, and asks Jev a Choice. Open-ended or freeform skills get a 'no prediction' report instead of a guess. The key is read from the environment at call time and never written anywhere. A Cursor skills mirror is included.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Does this skill reduce to a closed set of named outcomes? | yes · no |
| Choice | Which of the target skill's outcomes would it pick now? | SHIP · NEEDS_WORK · none_of_the_above |
How it works
- Resolve the target skill
- Gather read-only evidence into state
- Check with a Noul that outcomes are closed (threshold 0.65)
- Extract outcome labels from the skill and add none_of_the_above
- Ask Jev a Choice and report the pick
Strengths
- Avoids running slow decision-shaped skills
- Labels come from the target skill, never invented
- Refuses to guess for open-ended skills
Caveats
- Prediction, not execution; may differ from the real outcome
- Single prompt file without tests
- Requires a TypeSafe key
Getting started
Copy SKILL.md into your agent's skills directory, set TYPESAFE_API_KEY, then run jev-predict <skill>.At a glance
- Type
- Agent skill
- Status
- prototype
- Runtime
- Agent skill (SKILL.md)
- Has code
- yes
Use cases and integrations
How curated lists describe it
An Agent skill recipe that predicts another skill’s closed-set outcome from its rules and evidence.
logicrw/awesome-jev-projectsPredict another skill's next closed decision with TypeSafe Jev — without running that skill.
hellogumbo/awesome-jev
Frequently asked questions
What does it predict?
The named outcome another skill would choose, such as SHIP or NEEDS_WORK.
Does it run the target skill?
No. It gathers evidence and asks Jev to pick from the skill's own outcomes.
What if the skill is open-ended?
It reports no prediction instead of guessing.
Related projects
Self-hosted Jev MCP server for Claude Code and Codex with confidence gating, cost guardrails, caching and decision logs.
MCP serverbetaPythonKey-locked Netlify proxy that serves Jev through Netlify AI Gateway credits, plus an MCP server for Claude Code, Codex and Claude Desktop.
MCP serverbetaPythonSkill and MCP service that audits an app for AI calls that are really decisions and shadow-tests migrating them to Jev.
MCP serverbetaTypeScriptRust CLI for Jev that lints typed question sets offline, then sends them and prints probabilities, labels and scores as JSON.
CLIbetaRustPython router that has Jev classify Grok Bot requests before costly work, returning actions like reuse_cache, stop_retry or ask_human.
Agent skillprototypePython- Building with Jev★ 155
Agent skill for writing and improving programs that call Jev: question design, state structure, thresholds and diagnosing wrong answers.
Agent skillproduction