herdr-pr-flow
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes the herdr CLI and bundled Python scripts. It enforces a strict blocklist of dangerous commands (e.g., sudo, curl, rm) for project-specific checks to prevent unauthorized system modifications.
- [DATA_EXFILTRATION]: File access is managed via snapshot.py, which implements robust path validation to prevent traversal and explicitly forbids the inclusion of sensitive files or directories such as SSH keys and environment variables in task snapshots.
- [INDIRECT_PROMPT_INJECTION]: The orchestration logic processes data from multiple agents and mitigates injection risks by treating sub-agent output as untrusted. The HTML reporter performs comprehensive escaping and validates all participant data against frozen records.
- [DYNAMIC_EXECUTION]: The reporting utility dynamically loads the local configuration validation script to verify task metadata. This execution is restricted to the skill's own trusted files and does not incorporate unvalidated external input.
Audit Metadata