learning-workspace
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes local Python scripts (scripts/workspace_scaffolder.py and scripts/review_scheduler.py) to manage study workspaces and schedules. These scripts are executed locally by the agent to perform file I/O and process JSON data.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a potential surface for indirect prompt injection because it ingests untrusted data from user-managed files which are then used in prompts. Ingestion points: Untrusted data enters the agent context through sample_workspace_spec.json, learning-record.json, and generated lesson files in the lessons/ directory. Boundary markers: Data is delimited by standard Markdown and JSON structures, but the skill lacks explicit markers or instructions to ignore embedded commands within user-provided text fields. Capability inventory: The agent is empowered to execute the provided Python scripts and write files to the local file system. Sanitization: There is no evidence of sanitization or escaping of user input before it is interpolated into the Markdown templates used by the agent.
Audit Metadata