0153GitHub
super-jev
A small decision-to-action harness for Jev.
Kevthetech143/super-jevREADME ↗
# Super Jev [](https://github.com/Kevthetech143/super-jev/actions/workflows/test.yml) One judge (Jev) between your agent and your data: the agent asks in its own words, Super Jev finds the file, checks the claim, permits the action, and remembers what you approved. Your agent stays responsible for the answer. ## Vision - One judge sits between your agent and your data: the agent asks in its own words, Jev finds the file, checks the claim, permits the action, and remembers what you approved. Your agent stays responsible for the answer. - Agents burn whole LLM turns on lookups, re-hunt the same answers daily, and state things the files never said. Jev is a fast, cheap judge for yes/no and which-one questions; the LLM keeps the writing. - The daily loop: ask → read the top file → answer → approve / miss / add. Connectors are how your data gets in; the cache fills only from your own approvals — nothing is cached that you did not approve. - 1.0 promises the proven core: skill search, file navigate, connect + bulk prepare, check gate, permit gate, the harness loop. The core works well when it works; edges still want an agent in the seat — see KNOWN-QUIRKS.md and AGENTS.md. - 1.0 does not promise unattended answering, semantic cache matching, automatic sync, or live browsing. Next: auto-catch, recipes, a Jev-decided browser driver — each ships only after its own live bench. ## Quick start People: [docs/GETTING-STARTED.md](docs/GETTING-STARTED.md) — the full operating manual. Agents: [AGENTS.md](AGENTS.md) — the numbered path and daily loop, plus [wire-into-claude-code](docs/wire-into-claude-code.md) for the retrieval rule card and Stop-hook claim gate. At a terminal: `npm run jev` opens the app ([docs/jev-cli.md](docs/jev-cli.md)). No key yet? `npm run demo` runs the loo