ai-coding-agent-guardrails

Pass

Audited by Gen Agent Trust Hub on Mar 27, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Downloads the git-secrets tool from the official Amazon Web Services (awslabs) GitHub repository.
  • [EXTERNAL_DOWNLOADS]: References and integrates trusted security tools and images, such as Semgrep, TruffleHog, and official Ubuntu Docker images.
  • [COMMAND_EXECUTION]: Provides scripts for system-level configuration including user creation (useradd), filesystem access control (setfacl), and firewall rule configuration (iptables).
  • [COMMAND_EXECUTION]: Uses standard development and CLI utilities such as npm, pytest, git, and the GitHub CLI to manage CI/CD pipelines and auditing.
  • [PROMPT_INJECTION]: Processes AI-generated content (code, branch names, and PR descriptions), which serves as an indirect prompt injection surface.
  • Ingestion points: scan_agent_output.py (file content reading) and .github/workflows/agent-pr-review.yaml (PR metadata and body).
  • Boundary markers: Recommends defining constraints in CLAUDE.md, .cursorrules, and codex.md files.
  • Capability inventory: setup-agent-workspace.sh (system permissions), agent-network-controls.sh (egress filtering), and run-agent-sandbox.sh (container execution).
  • Sanitization: Employs regular expressions to identify and block potential secrets, although it does not perform content sanitization or escaping.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 27, 2026, 02:05 PM