Organization repository on GitHub
A Stop hook that stops your coding agent from stopping too early. Plain-language rules, judged by jev.
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Tools & CLIs
Limpet
A Stop hook that keeps coding agents from quitting early: rules written in plain language are scored by Jev in about 0.7 seconds, and a violated rule sends the agent back to work instead of letting it stop.
- Stars
- 6
- Language
- Python
- License
- MIT
- Reading time
- 4 min
- Published by
- Community
- Author
- noplan-inc
- Last verified
- Sep 2026
About this project
Your agent stops. limpet doesn't let it. Coding agents stop early. "I'll wait for CI." "Shall I run it?" "Please ask the team about this." You type do it. Then again. Then again. limpet is a Stop hook. You write rules in plain language. Every time the agent is about to stop, jev scores the stop against every rule in 0.7 seconds, and if a rule is violated the agent is sent back to work instead of stopping. It clings to the task like a limpet to a rock.
Quoted from the project's README. Read the full README.
Highlights
- Each stop sends the last three messages, this turn's tool calls, and the final message, and Jev answers one yes/no question per rule in parallel.
- A violated rule exits 2 with a one-line reason; the second stop of the same chain is always allowed through, so the agent is pushed back at most once.
- One Python file using only the standard library; a stop costs about 1,000 to 2,000 input tokens, roughly a hundredth of a cent.
- Installs as a Claude Code or Codex plugin, or as a Stop hook for any agent with Claude-style hooks, with the key stored in the OS keychain.
- suggest and calibrate read local transcripts to draft rules and report AUROC; failures exit 0 silently so a hook never blocks work.
Quickstart
bash
claude plugin marketplace add noplan-inc/limpet
claude plugin install limpet@limpetWatch out
MIT-licensed. Needs Python 3.9+ on PATH and a TypeSafe or Vercel AI Gateway key; tested on macOS and Linux, and suggest and calibrate read transcripts of every Claude Code project and Codex session on the machine.
Questions about this entry
Short answers drawn from the linked project or page.
No. The second stop of the same chain is always allowed through, so the agent is pushed back at most once per stop; if it still disagrees it says why in one line and stops. False positives cost one sentence, while missed early stops cost you a round trip, so the guidance is to tune for recall rather than precision.
The project calls it a cheap nudge, not a wall. On its own transcripts, AUROC runs about 0.57 to 0.64 for most rules, and at a threshold blocking 5% of fine stops it catches 5% to 12% of bad stops of that type, one to two and a half times random blocking. A rule below 0.55 is left out of the blocking list. The labels come from the same model, so they are noisy and the numbers are a floor.
No. Without the block environment variable set it runs in shadow mode: every stop is scored and logged to ~/.limpet/log.jsonl and nothing is blocked. calibrate scores your current rules.md against past stops and prints a threshold per rule, which you copy into the env block of settings.json or into ~/.limpet/env.
The hook itself only sends the last 3 messages, this turn's tool calls and the final message to the model, at 1,000 to 2,000 input tokens per stop. The optional suggest and calibrate commands read Claude Code and Codex transcripts across every project and session on the machine; 1,500 stops took about 100 seconds and 15 cents. Failures exit 0 silently so a hook never blocks work.
It reads the last 3 messages, this turn's tool calls and the final message, then Jev answers one yes/no question per rule in parallel in about 0.7 s; a violation exits 2 and sends the agent back to work.
At most once; the second stop of the same chain is always allowed through, so if the agent disagrees it says why in one line and stops.
Claude Code: /plugin marketplace add noplan-inc/limpet then /plugin install limpet@limpet. Codex: codex plugin marketplace add noplan-inc/limpet then codex plugin add limpet@limpet, and trust its Stop hook.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
YouTube: Ray Amjad puts Jev inside an agentic coding loop and looks at what it costs to run.
YouTube: RoboNuggets puts Jev to work inside Claude Code.
X: Decision outputs in Grep
Hacker News: jevmem keeps Claude Code project memory by asking Jev which turns are worth saving.
avinashjetwani61 pointsJevmem – automatic project memory for Claude Code, built on JevRead the thread on Hacker NewsMore Jev resources like this
A cross-platform memory layer for AI coding agents: conversations from Claude Code, Codex, DeepSeek Harness, OpenClaw, and OpenCode become Markdown memories indexed in Milvus, with optional Jev reranking through the TypeSafe API.
Jev Tools & CLIs#community#python#agents
Community
zilliztech/memsearchPython
zilliztech/memsearchPythonA collection of Jev demos, workflows, and agent skills for coding agents: five installable skills and 108 scenarios, installed by pointing Codex, Claude Code, or OpenCode at an agent prompt.
Jev Tools & CLIs#community#python#skills
Community
wuyoscar/jev-skillPython
wuyoscar/jev-skillPythonA collection of 26 production-ready agent skills for Claude Code, Cursor, Kiro, Windsurf, and OpenCode; four skills call TypeSafe Jev for calibrated Score and Noul judgments and fall back to heuristics when it is unavailable.
Jev Tools & CLIs#community#python#skills
Community
fabricioctelles/skillsPython
fabricioctelles/skillsPythonA Jev-guided context compaction library for coding agents: instead of summarizing old context, Jev decides which tool calls and results still matter while user and assistant text is kept verbatim.
Jev Tools & CLIs#community#typescript#compaction
A security hook for coding agents that asks Jev three typed questions before every tool call and scans tool results for prompt injection, with adapters for Claude Code, Codex, Copilot CLI, Gemini CLI, Cursor, pi, OpenCode and ACP.
Jev Tools & CLIs#community#javascript#agents
Community
leepokai/jev-guardJavaScript
leepokai/jev-guardJavaScriptAn Agent Skill that sends closed coding-agent judgments to TypeSafe Jev and acts on the probabilities, asking for yes/no, choice, and score answers in about 250 ms while the main coding model keeps writing code.
Jev Tools & CLIs#community#python#skills
Community
HyunjunJeon/jev-judgmentPython
HyunjunJeon/jev-judgmentPythonFrom 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.