planning-with-files
Pass
Audited by Gen Agent Trust Hub on Aug 22, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill processes project-specific planning files that could potentially ingest untrusted external content from research or tool outputs. It mitigates this surface by implementing multiple defense layers: delimiter-based data framing (using per-session nonces), explicit instructions to the agent to ignore any embedded instructions within the plan data block, and a SHA-256 attestation mechanism to verify plan integrity before injection.
- [COMMAND_EXECUTION]: The skill utilizes lifecycle hooks to execute local shell and PowerShell scripts for plan management and task completion gating. These scripts include robust security features such as path canonicalization to prevent directory traversal and concurrency locking to ensure data consistency during parallel execution.
- [COMMAND_EXECUTION]: On Windows systems, the skill invokes PowerShell with the
RemoteSignedexecution policy to run its internal management logic. This configuration is used exclusively for the skill's own local scripts and does not involve persistent modifications to system-wide security settings.
Audit Metadata