Releases: aerf-spec/agentmint-python
v0.1.0 — Runtime enforcement for AI agent tool calls
First public release. pip install agentmint
Shield — 23 compiled regex patterns scan tool inputs and outputs for PII, secrets, prompt injection, encoding evasion, and structural attacks.
Scoped delegation — Glob-style permissions enforced per action. Child agents get the intersection of parent scope — never more authority than the parent.
Circuit breaker — Per-agent sliding window rate limiter. Closed → half-open → open.
Session enforcement — Per-pattern counters with configurable escalation and deny thresholds. deny_after overrides in_policy to False.
Cryptographic receipts — Ed25519 on every allow and deny. SHA-256 hash chain. Optional RFC 3161 timestamps. Export a zip, verify with openssl.
JSONL sink — SIEM-compatible audit log.
184 tests. 2 dependencies. MIT licensed.