Agent Security Harness

Version updated for https://github.com/msaleme/red-team-blue-team-agent-fabric to version v4.1.0.

  • This action is used across all versions by ? repositories.

Action Type

This is a Composite action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

Agent Security Harness is a GitHub Action designed to conduct automated security testing of AI agent protocols, operational governance, and decision-making mechanisms. It provides 440 executable security tests across multiple layers, including wire-protocol integrity, session state validation, and decision governance, to identify vulnerabilities such as spoofing, capability escalation, and unsafe agent behaviors. This tool enables organizations to proactively simulate adversarial scenarios, ensuring compliance with security policies and robustness against advanced threats.

What’s Changed

What’s Changed

Full Changelog: https://github.com/msaleme/red-team-blue-team-agent-fabric/compare/v3.10.1...v4.1.0## v4.1.0 — Compliance Evidence

440 tests, 31 modules. Transforms the harness from a security testing tool into a compliance evidence platform.

New Features

FeatureDetails
AUROC metricsPer-module detection effectiveness scoring via trapezoidal ROC
EU AI Act crosswalk16 controls mapped across Articles 9, 10, 13, 14, 15, 17, 27, 50, 72
ISO 42001 crosswalk15 controls mapped across Clauses 5-10 + Annex A
Kill-switch testsIR-009 to IR-012: termination latency, post-kill prevention, bypass resistance, graceful degradation (CA SB 942, EU AI Act Art 14)
FRIA evidence6-category Fundamental Rights Impact Assessment output (EU AI Act Article 27)
Watermark testsWM-001 to WM-005: presence, forgery resistance, stripping resistance, extraction resistance, multi-language (EU AI Act Article 50)
Compliance reportOne-command auditor-ready HTML: python scripts/compliance_report.py --report results.json --output report.html --framework all --fria
Simulate mode expansionMCP (14), A2A (13), Identity (12) harnesses now support --simulate

Quick start

pip install agent-security-harness==4.1.0

# Run tests
agent-security test mcp --url http://your-agent --simulate

# Generate compliance report
python scripts/compliance_report.py --report results.json --output compliance.html --framework all --fria

Stats

  • 440 unique test IDs across 31 modules
  • 19/19 pytest passing
  • 31 framework controls mapped (EU AI Act + ISO 42001)
  • Independent audit R33: 7 findings, all resolved
  • Simulate coverage: 52% of harnesses (12/23)

Breaking changes

None. Fully backwards compatible with v3.10.x.

Full changelog: https://github.com/msaleme/red-team-blue-team-agent-fabric/compare/v3.10.1...v4.1.0

What’s Changed

Full Changelog: https://github.com/msaleme/red-team-blue-team-agent-fabric/compare/v3.10.1...v4.1.0