User repository on GitHub
TypeSafe Jev hooks for Claude Code, Codex and pi: per-turn skill suggestion and subagent model routing
- Forks
- 1
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Agents & MCP
Jev Agent Hooks
Hooks that put Jev in two spots of a coding agent's loop, picking which skill a turn needs and which model tier a subagent gets, with deterministic rule gates in front and fail-open behaviour everywhere else.
- Stars
- 2
- Language
- JavaScript
- License
- MIT
- Published by
- Community
- Author
- onlyjq04
- Last verified
- Oct 2026
About this project
Hooks that put TypeSafe Jev in a coding agent's loop, picking which skill to load for a turn. One shared implementation runs on Claude Code, Codex, and pi. A rule gate for subagent routing, which does not call Jev, ships alongside. Every hook fails open. A Jev outage, a missing API key, or a bad response lets the turn through. hooks/jev-skill-suggest.mjs runs on every user prompt, follows the skill-suggestion cookbook, and makes two Jev requests:
Quoted from the project's README. Read the full README.
Used inJev for Software & DevTools
- Languages
- JavaScript
Highlights
- Skill suggestion makes two Jev requests per turn: a Choice over the whole roster, then a re-rank of the top three using SKILL.md bodies.
- Routing is layered: a rule gate pins aliases and preset pairings, and the Jev gate only asks what each host's policy leaves open.
- A denial fires at most once per session, agent and tier, so resubmitting the same call unchanged goes through and a wrong answer cannot trap you.
- install.sh symlinks files into host directories and edits no settings file; you merge the two JSON snippets into your own config yourself.
Watch out
The thresholds that decide everything are constants at the top of each hook file, so tuning them means editing source rather than config. The offline evaluation is a replay of 34 real turns and 20 historical dispatches from one person's setup, and the README notes the label for a correct skill is whatever the agent loaded at the time, which is not ground truth. Codex encrypts `spawn_agent.message` in some builds and those payloads are skipped rather than judged. Everything in a Jev state goes to api.typesafe.ai, including prompts, agent descriptions and skill descriptions. Denial messages ship in Chinese. 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.
That is the cost the README reports for the suggester against the live API, roughly 1 second offline at p50, because it makes two round trips. The dispatch gate is about 400 ms. Both are per turn, not per tool call, and the suggester is the expensive one. If that is too much, `JEV_SKILL_SUGGEST=shadow` logs the pick without injecting anything, and `off` turns it off.
The design assumes it will eventually be wrong and bounds the damage. Every hook fails open, so an outage, a missing key or a bad response lets the turn through, and a denial fires at most once per session, agent and tier, so the identical resubmission goes through. The rule layer denies things that are structurally wrong regardless of the model, such as a concrete model id where the Agent tool only accepts aliases.
The README is explicit that they come from replaying real sessions rather than a benchmark, that they are small, and that they describe one person's skill set and habits. It also reports where the fit threshold did the work and where the three-Noul gate barely separated anything, and names two failure modes, including near-duplicate skills splitting each other's probability. Read them as a debugging trace, not a performance claim.
The README is plain about it: everything in a Jev state goes to api.typesafe.ai, and the hooks send prompts, agent descriptions and skill descriptions. It says not to put secrets in an agent prompt. The CLI log records question ids and answers, never the state that was sent, and every hook appends one JSON line per call under ~/.claude/logs/.
It denies a subagent with no explicit model, a model outside haiku|sonnet|opus|fable, a concrete model id instead of an alias, or a preset agent paired with the wrong model. The Agent tool only takes aliases, so versions are pinned once in settings env.
mech is haiku at low effort, bulk is sonnet at medium for executing a written-down approach, deep is opus at xhigh, and oracle is fable at xhigh as a read-only advisor. The rule gate enforces the pairing.
Requires Node 20 or newer and TYPESAFE_API_KEY. Run ./install.sh, which symlinks files into the Claude, agents, local bin and pi directories without editing any settings file, then merge config/claude-settings.snippet.json or config/codex-hooks.snippet.json and trust the file in Codex.
More Jev resources like this
4GitHub stars
OpenClaw plugin adding a jev_route tool that classifies task type, complexity and risk with TypeSafe Jev, then maps the result to a route through testable JavaScript policy.
Jev Agents & MCP#community#awesomejev#catalog-import
Community
trietphan/jev-clawJavaScript
trietphan/jev-clawJavaScript2GitHub stars
An MCP server packaged as both a Codex and a Claude Code plugin, exposing eight read-only tools that turn Jev probabilities into proceed/review recommendations for routing, ranking, extraction and claim checks.
Jev Agents & MCP#community#awesomejev#catalog-import
Community
E-FL/typesafe-as-a-judgeJavaScript
E-FL/typesafe-as-a-judgeJavaScript7GitHub stars
A local MCP server that gives Codex a few cited excerpts instead of a whole noisy search result, with a deterministic default selector that needs no Jev account or key.
Jev Agents & MCP#community#jevbest#catalog-import
Community
suenot/codex-jev-routerJavaScript
suenot/codex-jev-routerJavaScript2GitHub stars
A Node CLI that reads a folder of Word, Excel, PowerPoint, PDF, email, CSV, HTML, code and text files, classifies each with Jev on OpenRouter, and writes a catalog, an INDEX.md and an offline report. Originals are never moved.
Jev Agents & MCP#community#awesomejev#catalog-import
Community
nexibeo/jev-organizeJavaScript
nexibeo/jev-organizeJavaScript25GitHub stars
Agent-oriented playbook for using Jev as a cheap, fast decision layer that triages, classifies, scores, and gates work before expensive browser, research, retry, or subagent steps.
Jev Agents & MCP#community#awesomejev#catalog-import
Community
Bodila51/muse-jev-playbookPython
Bodila51/muse-jev-playbookPython14GitHub stars
Opt-in Hermes plugin that routes each turn to one skill: a wide Jev Choice plus request checks gate a narrow fits round, injecting a single ignorable hint line.
Jev Agents & MCP#community#jevbest#catalog-import
From the Jev community
Posts from builders shipping with Jev right now.