jev
Skill2 starsChanged 9 days agoMITView on GitHub
Conventions for the Jev decision tools. Use for non-biased technical decisions (jev_decide), pre-execution command screening (jev_action_gate), git patch reviews (jev_review), final acceptance gating (jev_gate), claim verification (jev_verify), prompt injection screening (jev_screen), and calibrated truth probabilities (jev_noul).
- Deletes or force-pushes
33 other projects use a version of this.
Adopted by
Most recently taken up by
- fralapo/awesome-agent-skillstheir own version, first seen Nov 7, 2026
- Lu1sDV/skillsmdtheir own version, first seen Nov 4, 2026
- josecabralf/skillstheir own version, first seen Nov 1, 2026
- dagonet/claude-code-toolkittheir own version, first seen Nov 1, 2026
- connorch/skillstheir own version, first seen Oct 28, 2026
- GastonGelhorn/jevmatetheir own version, first seen Oct 25, 2026
Used with
Projects with it also use review.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
SentrySkillreview
Fixture review skill.
Also in 5 projects that use this
Other versions33 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 33
- wuyoscar/jev-skill★ 669same name
- joetawil7/first-pass★ 128same name
- FrancoisChastel/jev-code★ 44same name
- BorisLeMeec/jev★ 38same name
- flc1125/skills★ 38same name
- pragma-sh/pragma★ 25same name
- ciberjohn/Hermes-Skills★ 21same name
- ratacat/ratacats-skills★ 9same name
- Vincent-the-gamer/skills★ 9same name
- AlexPEClub/Jev-Model-Router-Claude-Code★ 8same name
23 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
2 rules. 0 are common, 1 is distinctive, and 1 is very specific to this project.
0
Common
Most of the 29 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 16
Do NOT call Jev on every micro-edit, file read, or routine thought.
- Very specific: line 59
Anti-Fake Done: Never invent test logs or mock output to pass jev_gate.
“Common” is measured: kept by at least 6 in 10 of the 29 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Jev Decision Tools
Fourteen tools built on TypeSafe's Jev model. They return typed judgments, ordinal rubric scores, and calibrated probabilities in ~400–800ms. The tools advise; you enforce engineering policy.
The Three-Gate Cadence (Preventing Agent Paralysis)
Do NOT call Jev on every micro-edit, file read, or routine thought. Follow the Three-Gate Cadence:
[Agent Thinking / Planning]
│
▼
[2+ Non-Trivial Architectural Choices?] ──(Yes)──► Call jev_decide (once)
│ (No)
▼
[Agent Writes / Edits Code] (Run uninterrupted)
│
▼
[About to Run Destructive Shell Command?] ─(Yes)──► Call jev_action_gate (rm, drop, git reset --hard)
│ (Safe / Read-only)
▼
[Run Real Verification Checks / Tests] (bun test / npm test / cargo test)
│
▼
[Definition of Done / Task Completion] ──────────► Call jev_gate (ONCE on cumulative diff vs test logs)Tool Selection Matrix
| Tool | When to USE | When to SKIP |
jev_decide | Facing 2–6 architectural alternatives with empirical evidence and trade-offs | The choice is routine, trivial, or already specified by user |
jev_action_gate | Before running potentially destructive commands (rm, drop table, git reset --hard) | Routine read-only checks (ls, git status, cat, reading files) |
jev_gate | Calling a task done: evaluating git diff PLUS developer claims vs test logs | Routine work in progress where tests have not been executed |
jev_review | Scoring a git diff on correctness, spec match, test gap, and blast radius | No code diff exists (use jev_verify) |
jev_verify | A report, PR brief, or summary makes claims with cited document text | No evidence document exists to cross-check against |
jev_screen | Placing fetched web text or pasted untrusted content into context window | Text is from trusted local project files |
jev_extract | Extracting structured atoms (prices, versions, dates, IDs) with zero hallucination | Free-form prose that cannot be matched by candidate regex |
jev_audit | Verifying model-extracted key-value pairs against source text | Value was already bounded verbatim by jev_extract |
jev_find | Finding the single best hit by semantic meaning (files, lines, notes) | Exact regex or keyword match decides it deterministically (use grep/rg) |
jev_rerank | Full ordering matters across a retrieval set | Only the single best hit is needed (use jev_find) |
jev_compare | Comparing two passages for factual agreement, conflict, or different facts | Relationship between passages is already known |
jev_noul | Calibrated probability (0.0 to 1.0) for a standalone proposition | Testing claims strictly against supplied evidence (use jev_verify) |
jev_judge | Universal flexible decision primitive (Yes/No, choice, ordinal scale) | A specialized domain tool exists |
jev_judge_batch | Concurrently evaluating multiple independent items in a queue | Single item judgment (use jev_judge) |
Operational Policies & Fail-Closed Rules
- Verify Before Presenting: Correct contradicted claims. Add real evidence logs for unsupported claims.
- Anti-Fake Done: Never invent test logs or mock output to pass
jev_gate. A contradicted claim (verdict: "contradicted") is an immediate stop and escalate, not a footnote. - Decide Once: An
escaped: truerecommendation (ask_user,investigate,none) means pause and gather missing facts or ask the user—do not rephrase and retry. - Documentation Diffs: For non-code changes (markdown, docs, styling), test coverage gap is treated as negligible (0) so documentation updates are not falsely rejected.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

