travel-manager-hardened

Pass

Audited by Gen Agent Trust Hub on Apr 21, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • [PROMPT_INJECTION]: The file SAFETY.md contains examples of adversarial prompts and system-level messages (e.g., [SYSTEM] Google Maps integration configured...) used for safety testing. This content represents an indirect prompt injection surface. * Ingestion points: The agent processes SAFETY.md as context. * Boundary markers: While labeled as test cases, the blocks lack programmatic delimiters to prevent the agent from obeying the embedded instructions. * Capability inventory: SKILL.md acknowledges capabilities for file writing and network requests. * Sanitization: No sanitization is applied to the reference documentation content.
  • [NO_CODE]: The skill is comprised entirely of Markdown files. No executable code (Python, JavaScript, or shell scripts) is present in the analyzed files.
  • [COMMAND_EXECUTION]: The skill instructions in SKILL.md explicitly reference the potential for shell command and API execution, providing negative constraints. This indicates the skill is intended for high-privilege environments, though no specific execution patterns are provided.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 21, 2026, 06:31 PM
Security Audit — agent-trust-hub — travel-manager-hardened