HN Showcase
Back to browse
GitHub Repository

Core execution boundary engine (decision + proof + ledger)

1 starsPython

Fail-closed execution guard for AI agents (Python, pip installable)

by echo_os·Mar 1, 2026·1 point·1 comment
Visit ProjectView on HN

AI Analysis

●●SolidSolve My Problem

Fail-closed execution guard with signed proofs—but risk scoring lacks published methodology or benchmarks.

Strengths
  • •Fail-closed default (no policy = DENY) is the right security posture for agent execution gates.
  • •Severity-adaptive thresholds adjust dynamically based on system state, not static policy.
  • •Signed boundary proofs prevent agent retry-loops past denial decisions.
Weaknesses
  • •Risk scoring algorithm not documented; unclear how 'wire_transfer amount=50000' scores as 40/100 vs 80/100.
  • •No published benchmarks, false-positive/negative rates, or real-world deployment evidence.
Category
Security
Target Audience

Enterprise teams running autonomous AI agents in high-stakes domains (finance, healthcare, content moderation).

Similar To

Anthropic Constitutional AI · LangChain guardrails · DeepEval validators

Similar Projects

Security●●●Banger

NPM install is a security hole, so we built a guard for it

Blocks malicious packages at install-time before AI agents execute them on your machine.

Big BrainSolve My Problem
Sahil121
102mo ago