add-seed-data
Pass
Audited by Gen Agent Trust Hub on Nov 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill includes strong, explicit prohibitions against exfiltrating data or using private secrets. It mandates the use of synthetic or explicitly authorized data and forbids adding network access, capture scripts, or live-mode migrations.
- [UNVERIFIABLE_DEPENDENCIES]: The instructions acknowledge dependencies on other local skills (product-design, setup-skills, create-prototype) but strictly forbid automatic installation or setup, requiring the agent to stop if files are missing rather than triggering a download.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface because it processes and edits external data files (fixtures/scenarios).
- Ingestion points: Reads canonical fixtures, raw types, and data-to-view assembly files from the local repository.
- Boundary markers: No specific delimiters for data are defined in the instructions.
- Capability inventory: The skill has authority to modify local raw input files and run consumer-owned validation checks.
- Sanitization: Mitigated by instructions requiring format/type validity checks, uniqueness validation, and referential integrity checks before changes are committed.
Audit Metadata