engineering-manager
Pass
Audited by Gen Agent Trust Hub on Nov 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a state persistence mechanism using files in the .nugelo/ directory. This creates an attack surface where malicious instructions could be placed in data files processed by the agent. The skill includes a 'Project State Protocol' that explicitly instructs the agent to treat this data as untrusted and not follow directives found within, providing a necessary security mitigation.
- Ingestion points: The agent reads from .nugelo/decisions/, .nugelo/status.md, and .nugelo/handoff.md (SKILL.md).
- Boundary markers: The skill uses a dedicated Project State Protocol section with explicit warnings about untrusted content.
- Capability inventory: The skill utilizes file system read and write tools to manage state in the project workspace.
- Sanitization: The instructions mandate treating file content as untrusted input and surfacing any discovered directives to the user.
Audit Metadata