Overview & result
Pi-Heed keeps the user's changing instructions as structured policy. Jev classifies policy updates from conversation; deterministic code checks side-effecting Pi tool calls against the current policy before they run.
From the original post
Your agent understood your instruction. pi-heed makes sure it still remembers.
Runtime constraints for the pi coding agent: every side-effecting tool call is checked against what you said, before it runs.
How Jev fits in the loop
- Classify the user's policy additions, exceptions, and revocations with Jev.
- Store the resulting rules as structured session policy.
- Check each side-effecting Pi tool call against that policy in code before execution.
- Block, require confirmation, or permit the proposed action according to the active rule.
How to reproduce
- Clone Nyarlathoteppppp/pi-heed and hook it into the Pi agent runtime.
- Trigger an agent workflow that attempts file writes, bash commands, or network requests.
- Inspect how Jev classifies conversational policy updates, then test the deterministic gate against proposed side-effecting calls.
Why this build matters
Solves the primary trust barrier in autonomous coding agents by enforcing automated, low-latency oversight over dangerous tools.
Limitations
- Relies on clear natural language descriptions of the tool's intended effect.
- Cannot analyze deeply obfuscated shell scripts or complex binary executions.
- Requires user confirmation for low-confidence boundary decisions.