agent_guardrails

Self-healing multi-agent guardrail hub — four specialised agents, 44+ OPA rules, auto-fix loop with escalation

ACTIVE R&D

Self-healing guardrail hub running on a personal GitHub repo. Benchmark Agent orchestrates Guardrail Enforcement (34 Harness rules), Pi Guardrail (10 Pi workflow rules), and a weekly Research Agent — gating every PR before merge, auto-fixing up to three rounds, then escalating to a tracked GitHub Issue. 44 rules, 7 test files, 100% coverage. Ship faster.

Python
Bash
OPA/Rego
GitHub Actions
NVD/CISA/OWASP

Agents

Benchmark Agent
Hub & orchestrator
Single entry point for all developer interactions — code review, Q&A, triage.
Guardrail Enforcement Agent
OPA policy evaluation — 34 Harness rules across pipelines, connectors, delegates, code
Called by Benchmark Agent.
Pi Guardrail Agent
10 Pi workflow rules — bash security levels, code standards, workflow gates
Called by Benchmark Agent.
Research Agent
Weekly threat intelligence — NVD CVE, CISA KEV, OWASP; drafts new policy skeletons
Schedule: Mondays 08:00 UTC.

Workflow

Auto-fix committed to branch with PR comment; past limit → GitHub Issue opened with full loop history.

0stars0forks0watchers5open issues

main·0.15 MB·Open Policy Agent 68% · Python 32%

5ca7116ci: fix label creation failing due to missing git context before checkoutArif · 27d ago