datahub-connector-planning
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill researches source systems by fetching external web content (documentation, PyPI info, Docker Hub), which introduces a surface for indirect prompt injection where malicious instructions could be embedded in technical documentation.
- Ingestion points: Research phase in SKILL.md fetches content via WebSearch, WebFetch, and sub-agent tasks.
- Boundary markers: The skill explicitly instructs the agent to wrap untrusted research results in tags.
- Capability inventory: The skill has the ability to write a _PLANNING.md file and execute metadata-gathering shell commands (ls, find, grep).
- Sanitization: Explicit instructions are provided to ignore any embedded instructions in external content and extract only factual information.
Audit Metadata