User repository on GitHub
Fail-open Codex UserPromptSubmit hook that injects TypeSafe Jev pre-task routing metadata.
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Agents & MCP
Codex Jev Preflight
A Codex UserPromptSubmit hook that asks Jev for task_type, complexity, risk and execution_mode before a task starts and injects the result as advisory context, failing open on timeouts, quota errors and invalid responses.
- Stars
- 2
- Language
- Python
- License
- MIT
- Published by
- Community
- Author
- wellkilo
- Last verified
- Oct 2026
About this project
A fail-open preflight hook for every Codex task. Ask TypeSafe Jev for tasktype, complexity, risk, and executionmode before Codex starts working. The result is advisory, validated, and never blocks the task. Live demo · Quick start · How it works · Configuration · Security · 中文. If this project is useful, consider giving it a star ⭐.
Quoted from the project's README. Read the full README.
Used inJev for Software & DevTools
- Languages
- Python
Highlights
- Four routing fields are validated against their enums before injection, with unknown values collapsing to unknown.
- A persistent breaker stops calling after a confirmed exhausted quota instead of retrying on every prompt.
- Python standard library only, and the test suite runs fully offline without ever touching the live Jev API.
- Key entry uses hidden input with a 0600 config file, and the installer does not mark the hook trusted automatically.
Watch out
Every prompt sends the first 24,000 characters of the user prompt to a third-party endpoint, so sensitive text has to be filtered before it gets there — the README says so directly. The assessment is advisory context only and cannot override system instructions, developer instructions or an explicit user request, and Codex still decides for itself whether a task is answered, inspected, planned or clarified. Trust is opt-in: codex-jev-install only trusts the hook when passed --trust, and you need to restart Codex or open a new task afterwards. Tests never call the live API, so a green suite tells you nothing about your own quota or endpoint. The project states it is not affiliated with OpenAI, Codex, TypeSafe or Jev. 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.
It is fail-open by design. If Jev times out, hits a quota limit or returns something invalid, the hook falls through and Codex continues normally with no assessment block. The result is also only advisory context: it cannot override system instructions, developer instructions or an explicit user request. A persistent breaker adds a second layer by stopping calls entirely once a quota exhaustion is confirmed.
Yes, the first 24,000 characters of the current user prompt go to the configured TypeSafe endpoint. API keys are not logged and real env files are excluded by .gitignore, but the README's own advice is not to include sensitive data that should not go to a third-party service. You can point TYPESAFE_API_ENDPOINT somewhere else, or turn on JEV_HOOK_DEBUG_LOG while you work out what is being sent.
No, and that is listed as a feature under manual trust. The installer does not trust the hook automatically; you either approve it when Codex prompts or pass --trust to opt in explicitly, and installation preserves any existing hooks.json and AGENTS.md configuration. After installing you need to restart Codex or open a new task for the hook to be picked up.
TYPESAFE_API_ENDPOINT, JEV_MODEL, JEV_STATE_PATH and JEV_ENV_FILE cover it, and the lookup order runs from JEV_ENV_FILE down through .env beside the project source, .env in the working directory, $CODEX_HOME/jev.env which is normally ~/.codex/jev.env, and finally the XDG config path. Existing process environment variables win over all of them, and JEV_MODEL defaults to jev-latest.
task_type is answer, code_change, research, browser_automation, planning, conversation or other; complexity is trivial, simple, moderate or complex; risk is low, medium or high; and execution_mode is direct_answer, inspect_then_act, plan_then_execute or ask_clarification. Unknown values become unknown.
Run python -m unittest discover -s tests -v for the offline suite, then pipe a JSON payload with hook_event_name set to UserPromptSubmit into the hook module. The output should contain JEV PRE-TASK ASSESSMENT, and the tests never call the live Jev API.
Remove the persistent state file, or run python -c 'from jev_agent import get_default_client; get_default_client().reset()'. The breaker exists to stop calling Jev after a confirmed exhausted quota rather than retrying on every prompt.
More Jev resources like this
25GitHub 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-playbookPython23GitHub stars
Coding-agent plugin that routes prompt classification, subagent tiering, edit rule checks, and fast Jev-backed compaction through small typed judgments.
Jev Agents & MCP#community#jevbest#catalog-import
Community
0x7067/claude-jevTypeScript
0x7067/claude-jevTypeScript14GitHub stars
Polymarket bot that pairs a web-search researcher brief with Jev pricing, trading only where Jev probability clears the order book by a set edge.
Jev Agents & MCP#community#awesomejev#catalog-import
Community
markusbug/jevymarketPython
markusbug/jevymarketPython14GitHub 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
5GitHub stars
CLI that routes a task to one specialist skill from a Tink library using TypeSafe Jev, verifies the winner with tink mount, then prints it inline or as a mounted path.
Jev Agents & MCP#community#awesomejev#catalog-import
Community
jon-devlapaz/tink-routePython
jon-devlapaz/tink-routePython9GitHub stars
Hermes Agent plugin exposing Jev as four coding-workflow tools for parallel judgments, binary checks, option routing and rubric scoring, with a bundled question-design skill.
Jev Agents & MCP#community#jevbest#catalog-import
Community
ajensenwaud/hermes-jev-pluginPython
ajensenwaud/hermes-jev-pluginPythonFrom the Jev community
Posts from builders shipping with Jev right now.