DanielKillenberger/jev-predict-skill
User repository on GitHub
Predict another skill's next closed decision with TypeSafe Jev — without running that skill.
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Agents & MCP
Jev Predict Skill
A single-file host agent skill that reports what another skill's next closed decision would be, with a confidence and a probability over every outcome, by asking Jev to choose rather than by running the target skill at all.
- Stars
- 4
- Language
- HTML
- Published by
- Community
- Author
- DanielKillenberger
- Last verified
- Oct 2026
About this project
Predict another skill's next closed decision — without running that skill. jev-predict-skill is a host agent skill. Point it at any other skill and it reports the discrete outcome that skill would pick right now, with a confidence and a probability over every outcome. It gets there by asking TypeSafe Jev to make the pick, not by executing the target skill.
Quoted from the project's README. Read the full README.
Highlights
- Outcome labels are taken from the target skill itself, with none_of_the_above always added, so the set is never invented.
- A noul breakdown question gates the run at 0.65, and below that it reports that it cannot break the skill down and stops.
- It fails closed: a bad argument, an unbreakdownable skill, an auth or HTTP or parse error, or a none_of_the_above pick.
- Just a SKILL.md with no package and no build step; TYPESAFE_API_KEY is read at call time and never written to disk.
Watch out
It only suits skills that end in a predetermined, named set of outcomes, such as a review verdict, a phase choice or a router label; open-ended generation and freeform writing are explicitly out and return a no-prediction report rather than a guess. It predicts the next decision only, not a sequence of downstream picks. Two things to keep in mind about what it is doing: the target skill is never executed, so what reaches Jev is a compact read-only evidence set gathered from the files rather than the skill's real behaviour in a run, and the outcome set must come from the target skill's own definitions. The exact questions, thresholds, and HTTP and retry rules live in SKILL.md rather than in prose, and the skill is mirrored for Cursor under .cursor/skills. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
No, it never runs the target skill. It resolves the argument to a skill file, gathers the read-only evidence that the next decision depends on into Jev's state, asks whether the skill reduces to a closed set of outcomes, extracts the labels from the target skill, and asks Jev to pick one. You get a prediction of the verdict, not a rehearsal of the run.
It fails closed, and a stop is treated as a valid outcome. Below 0.65 on the breakdown question it reports that it cannot break the skill down; an auth, HTTP or parse failure produces a no-prediction line such as no prediction, assessment failed, 401 auth; and a none_of_the_above pick means no prediction either.
Only closed-decision skills, ones ending in a finite named set of outcomes like SHIP or NEEDS_WORK, a routing label, or approve and reject. The outcomes have to come from the target skill itself. Skills doing open-ended generation or freeform writing are out, and pointing at one gets a no-prediction report, not a guess.
There is no package and no build step. Copy the single SKILL.md wherever your agent looks for skills, or point the agent at the repository. Set TYPESAFE_API_KEY in the environment; the skill reads it at call time and never writes it into a file, a report or a commit.
The argument can be a skill file path, a directory containing a SKILL.md, a slash or skill name resolved under the usual skill roots, or pasted frontmatter. Invoke it as /jev-predict-skill, or as jev-predict <skill> with an argument like jev-predict /flow-next:impl-review.
A predicted label, a confidence figure, a probability for every extracted outcome including none_of_the_above, and the breakdown noul score saying whether the skill reduced to a closed set at all. Labels are read from the target skill and never invented, so an outcome the skill does not define cannot appear in the distribution.
Yes, SKILL.md is the full contract with the exact questions, thresholds, HTTP and retry rules, mirrored at .cursor/skills/jev-predict-skill/SKILL.md for Cursor. It also states the limit plainly: only the next closed decision is predicted, not a sequence of downstream picks.
More Jev resources like this
7GitHub stars
Portable agent skill that guides a coding agent through choosing, designing, and testing Jev uses, including when plain code is the better answer.
Jev Agents & MCP#community#jevbest#catalog-import
Community
karanb192/jev-architectHTML
karanb192/jev-architectHTML51GitHub stars
Local LinkedIn recruiting workspace on Browser Use's Jev Ultrafast: Jev screens titles, drives navigation, and assesses profile excerpts against your brief while a live browser stays visible; matches and evidence export as JSON.
Jev Agents & MCP#community#jevbest#catalog-import
Community
skeptrunedev/jev-recruiterPython
skeptrunedev/jev-recruiterPython33GitHub stars
Codex Skill plus stdlib-only Python CLI that recommends which installed skill to read next for a task, allowing no-skill or uncertain answers and never executing candidates.
Jev Agents & MCP#community#jevbest#catalog-import
Community
win4r/jev-skill-suggesterPython
win4r/jev-skill-suggesterPython72GitHub stars
MCP server over stdio that gives coding agents fourteen Jev judgment tools, turning evidence plus an enumerable question into auto, review, or escalate actions from calibrated probabilities.
Jev Agents & MCP#community#jevbest#catalog-import
Community
PyModel/jev-judge-mcpPython
PyModel/jev-judge-mcpPython21GitHub stars
Research runtime that pages a large logical tool and context space through a small bounded Jev decision surface with explicit approvals.
Jev Agents & MCP#community#jevbest#catalog-import
35GitHub stars
Rust terminal tool that asks Jev noul, choice and score questions about piped text or files and returns calibrated probabilities with stable exit codes for shell scripts, CI jobs and agents.
Jev Agents & MCP#community#jevbest#catalog-import
Community
shaharia-lab/jev-cliRust
shaharia-lab/jev-cliRustFrom the Jev community
Posts from builders shipping with Jev right now.