The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (such as organization names, assessment scores, and user notes) which can influence agent behavior or be reflected in exported reports.
Ingestion points: Data enters the system via record_responses_from_df and record_response methods in SKILL.md, as well as file imports (CSV, Excel, JSON) mentioned in instructions.md.
Boundary markers: The instructions and code do not implement explicit boundary markers or delimiters to separate user data from system instructions.
Capability inventory: The skill has file system write capabilities through the export_to_excel method in SKILL.md (using the filesystem permission declared in claw.json).
Sanitization: There is no evidence of input validation or sanitization for descriptive fields like notes or organization_name before they are processed and written to the filesystem.

digital-maturity-assessment