cross-conversation-project-manager

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill instructions define a workflow for maintaining project state in a designated local directory (/mnt/user-data/outputs/projects/). This behavior is consistent with the stated purpose of persistent project management and does not involve unauthorized file access or network activity.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied project names and conversation summaries to update internal tracking files. While this creates a surface for indirect instructions if a user provides malicious content as a project update, the risk is minimal as the agent's tools are focused on project reporting and file maintenance within a scoped directory.
  • Ingestion points: User-provided project names and conversation updates processed at runtime (SKILL.md).
  • Boundary markers: None explicitly defined in the instructions.
  • Capability inventory: Local file write operations within /mnt/user-data/outputs/projects/ (SKILL.md).
  • Sanitization: None specified for user-supplied project metadata.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 07:25 AM
Security Audit — agent-trust-hub — cross-conversation-project-manager