dbreunig/lunaroute-systemone-auto
User repository on GitHub
- Forks
- 0
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Agents & MCP
Jev Auto
An auto-mode safety monitor for the Pi coding agent: every tool call Pi is about to run gets one System One classifier question per rule, then allow, block or ask is decided in code. Built with DSPy against TypeSafe's Jev and exported to a Pi extension calling lunaroute/djev by default.
- Stars
- 4
- Language
- HTML
- Reading time
- 2 min
- Published by
- Community
- Author
- dbreunig
- Last verified
- Oct 2026
- Languages
- PythonTypeScript
- Platforms
- CLI
- Capabilities
- Open source
Highlights
- One Noul question per rule, decomposed from a monolithic classifier prompt: 1 HARD and 62 SOFT rules, plus consent checks.
- The README counts 127 questions in total, each asked separately instead of sending the whole prompt.
- Code, not the model, handles chained commands, encoded payloads and session-written files in jev_auto/state.py.
- HARD versus SOFT severity, consent and composition are computed in code by AutoModeMonitor.compose.
- Benchmarked in DSPy 3.4 against TypeSafe's Jev, then exported to the pi-system-one-auto TypeScript extension.
Watch out
README reviewed 2026-10-17; the digest excerpt ends mid-sentence at the question count, so the benchmark numbers, install steps and default endpoint credentials are not documented here, and no license is listed.
Questions about this entry
Short answers drawn from the linked project or page.
An auto-mode safety monitor for the Pi coding agent. Every tool call Pi is about to run is judged by a program that asks a System One classifier one question per rule, then decides allow, block or ask in code.
The program is written and benchmarked in DSPy 3.4 against TypeSafe's Jev in jev_auto/, then exported to a TypeScript Pi extension, the pi-system-one-auto package, which calls lunaroute/djev by default.
From auto_prompt.md's 1 HARD and 62 SOFT rules, one Noul per rule asks whether the action matches, each soft rule's consent bar becomes a consent Noul, and ALLOW exceptions are folded into the rules they narrow.
Chained commands, encoded payloads, executing session-written files and git status meta are handled in jev_auto/state.py, ported to the extension's source; HARD versus SOFT, consent and severity compose in AutoModeMonitor.
The README counts 127 questions in total, each traveling as a separate typed Noul question rather than one monolithic LLM prompt, which the README says follows the Jev skill's decomposition guidance.
Not fully documented in the digest excerpt: it targets the Pi coding agent and needs classifier access, but install steps, the lunaroute/djev endpoint requirements and key handling are cut off in the excerpt.
The GitHub user dbreunig publishes it; no license appears in the repository metadata. The repository was created 2026-10-05 and pushed the same day, with 4 stars when the sweep ran.
More Jev resources like this
A Pi extension that checks each agent reply against a list of skills: a classifier model, Jev by default, scores adherence from 0 to 1, and when a reply does not adhere the agent is prompted to edit it until it does.
Jev Agents & MCP#community#github#pi
Community
amalavet/pi-reply-guardTypeScript
amalavet/pi-reply-guardTypeScriptDevMortimer's Pi extension hands batched Choice, Score, and Noul judgments to Jev: a typesafe_evaluate agent tool, a terminal playground, and a typed API other extensions reuse, with daily request, token, and dollar caps.
Jev Agents & MCP#community#pi#typescript
Community
An experimental coding agent CLI built on opencode whose v2.0.0 architecture adds a JEV decision layer with five decision gates, shadow mode, and ECE calibration, plus a three-level RSI self-evolution loop and six Pi-framework extensions.
Jev Agents & MCP#community#github#typescript
Community
momozi1996/momo-codeTypeScript
momozi1996/momo-codeTypeScriptReusable project context and specialist workflows for the AI tools you already use: agents, skills, rules, and MCP tools for Claude Code, Codex, Copilot, and Cursor, plus Jev decision support that offloads bounded LLM decisions to save tokens.
Jev Agents & MCP#community#github#agents
Community
jmagly/aiwgTypeScript
jmagly/aiwgTypeScriptA free Claude Code skill that puts Jev in front of the expensive model: thirteen commands sort, check, score, gate and verify with Jev, so Claude reads only the unsure slice and writes the output. Needs python3 and an OpenRouter key.
Jev Agents & MCP#community#python#claude-code
Community
charlesdove977/claude-x-jevPython
charlesdove977/claude-x-jevPythonA very WIP tool for jev-ification from Harry Solovay: lessons from coding sessions become approved Markdown traits, and a configurable decision model evaluates code one trait at a time while an MCP server handles group-aware approval.
Jev Agents & MCP#community#github#typescript
Community
harrysolovay/traits-pocTypeScript
harrysolovay/traits-pocTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.