Recent agent platforms (e.g., Moltbook/Moltbot) showed how easily irreversible mistakes happen. This repo demonstrates a deterministic pre- and pre-execution gate that stops them.
Stop scams twice: block at the entrance (Pre-Guard) and right before execution (Tx-Guard), then save the reason as a single artifact (Chain).
- Verdict: PASS / DELAY / BLOCK
- Merge rule: severity = max, evidence = union
- Output: JSON artifacts (auditable, shareable)
Requires: Python 3.10+ (recommended)
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt 2>/dev/null || true