User repository on GitHub
Agent skill that spots bounded-judgment steps, tries a typed decision model (TypeSafe's Jev) first, and documents every attempt
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Agents & MCP
Decision First
Agent skill that pushes a coding agent to try Jev before it writes another regex or prompt, then writes down what happened. Four reference files, an eval set of 20 trigger prompts, and three stdlib Python scripts for running questions, scoring agreement against labels you already have, and keeping a decision lab.
- Stars
- 2
- Language
- Python
- License
- MIT
- Published by
- Community
- Author
- harrymunro
- Last verified
- Oct 2026
About this project
Teach your coding agent to try Jev before it writes another regex, and to write down what happened.
Quoted from the project's README. Read the full README.
- Languages
- Python
Highlights
- Installs as a skill in Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Pi or Hermes, and announces the candidate decision in one line first.
- Runs 5-20 real items through a stdlib runner that prints tokens, cost, p50 latency and agreement against any labels you already have.
- Every attempt is documented: declines get one line in a log, adoptions get a case directory with the questions verbatim and a run.sh.
- Names what Jev cannot do, counting, arithmetic, date comparison and wide technical option sets, and sends those parts to code.
Watch out
This is a prompt-and-procedure change rather than a model change, so whether the agent actually reaches for Jev depends on how much your agent ignores the skill. The durable value is the paper trail, not the routing. Whatever you put in state is sent to TypeSafe's API under your own key, so secrets and confidential text do not belong in a questions file. Tests run without network or a key, and the agreement figures it prints are only as good as the labels you already have. Independent project, not affiliated with TypeSafe AI. 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.
A nudge with a procedure attached. The skill tells the agent to spot a bounded judgment, announce the candidate in one line so you can refuse it, try Jev on a handful of real items, then log the attempt either way. There is no library to import: the scripts are plain Python over the HTTP API, and you can run ask.py, compare.py and lab.py by hand.
compare.py scores agreement per question against a labels file you already have and prints confusion tables plus the confident misses, which is what stops a plausible-looking rule from being integrated on a shrug. The skill asks for that check before adopting anything, and a proposal gets a case directory only if it survives it.
Whatever you put in state goes to TypeSafe's API with your key, and the README is direct about it: do not send secrets, and think before sending confidential text. There is no local inference path here, since the scripts are an HTTP client rather than a bundled model. If that matters, run the decision lab against data you are allowed to transmit.
The optional prompt_hook.py is a UserPromptSubmit hook that nudges when a prompt contains trigger vocabulary, and it never blocks and always exits 0. The default workflow is announce, test against ten decision shapes, try, measure, decide, document, report, rather than a hook on every message. The repo also ships 20 trigger prompts and a script for measuring how often the skill triggers.
SKILL.md holding the reflex and the procedure, four reference files covering decision shapes and question templates, launch-week field notes, the documented failure modes of jev-1.13, and an API cheatsheet. Alongside them sit assets/case-README.md as the case write-up template and an evals directory with 20 trigger prompts plus a script that measures how often the installed skill fires.
Export TYPESAFE_API_KEY and write your own questions.json, copying a template out of the shapes reference. Then ask.py --questions with either --state-text for one state or --items with --wrap for a JSONL batch, compare.py against a labels file, and lab.py to log a trigger or open a case directory. There are no dependencies beyond Python 3.11 plus, and the runner falls back to the system certificate bundle on Python builds shipped without one.
By default at ~/Workspace/decision-lab, or whatever directory $DECISION_LAB points at. Declines get a single line in a log; adoptions and experiments get a case directory holding the questions verbatim, a run.sh, results, and a how-to-reuse-this section written first, so reading the lab in three months shows where the approach actually paid off.
More Jev resources like this
51GitHub 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-suggesterPython32GitHub stars
Daily AI news pipeline that gathers RSS, arXiv, and social sources with staged agents and publishes a searchable web frontend.
Jev Agents & MCP#community#jevbest#catalog-import
Community
flyryan/ai-news-aggregatorPython
flyryan/ai-news-aggregatorPython72GitHub 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
18GitHub stars
Python CLI and MCP server for keyless web search over DuckDuckGo or self-hosted SearXNG, with an optional local Laya sieve that filters, reranks and stance-checks evidence before it reaches an agent.
Jev Agents & MCP#community#jevbest#catalog-import
From the Jev community
Posts from builders shipping with Jev right now.