Skip to content
System One

noplan-inc/limpet

Stop hook that keeps a coding agent from finishing early. You write completion rules in plain language, and every time the agent is about to stop, Jev scores the stop against each rule in 0.7 seconds and sends it back to work if a rule is violated. Works with Claude Code and Codex.

noplan-inc/limpet on GitHub

Stars
3
Forks
1
Language
Python
License
MIT
Last push
17 Sep 2026

Read from GitHub on . Counts change daily.

How it uses Jev

HERE = os.path.dirname(os.path.abspath(__file__))
HOME = os.path.expanduser("~/.limpet")
PROVIDERS = {  # url, model, yes/no question type, answer field
    "typesafe": ("https://api.typesafe.ai/v1/systemone", "jev-latest", "noul", "noul"),
    "vercel": ("https://ai-gateway.vercel.sh/v4/ai/evaluation-model", "typesafe-ai/jev", "boolean", "probability"),
}
TIMEOUT = 20  # jev call. Worst case before it: keychain 2 s + LIMPET_KEY_CMD 8 s => 30 s, the hook's budget
TAIL_BYTES = 512 * 1024  # the hook reads only the tail of the transcript
SCAN_BYTES = 4 * 1024 * 1024  # suggest/calibrate read this much per transcript
N_CONTEXT = 3  # previous messages sent as context
MAXLEN = 2000  # per message
MAX_TOOLS = 40  # tool calls of the current turn, most recent kept

# user lines that look human but are injected by the harness (collected from real transcripts)

View in limpet.py

Open on GitHub

Use casesLLM guardrails

More like this

Use cases this is tagged with