report
Pass
Audited by Gen Agent Trust Hub on May 10, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes untrusted finding details from local files and passes them to sub-agents.\n
- Ingestion points: Findings data is read from findings.md and included in the instructions for the report-writer and poc-builder agents.\n
- Boundary markers: Present. The skill uses "CRITICAL INSTRUCTIONS" and defined output schemas to maintain control over the agents' execution.\n
- Capability inventory: Includes the ability to execute local shell commands via uv run and write files to the project directory.\n
- Sanitization: Absent. The skill does not explicitly filter or sanitize the finding details before interpolation.\n- [COMMAND_EXECUTION]: Executes local Python scripts (dedup_findings.py, global_brain.py, statusline.py) located within the project's tools/ directory for operational tasks like finding deduplication and state syncing.\n- [SAFE]: Implements secure handling of identities by mandating the use of environment variable placeholders (e.g., ${HACKERONE_EMAIL_ALIAS}) in reports, effectively preventing the exposure of hardcoded credentials.
Audit Metadata