research-before-design
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to retrieve and analyze external content, including official documentation, technical articles, and open-source project repositories. This creates an attack surface where instructions embedded in these external resources (e.g., hidden HTML comments or documentation instructions) could attempt to influence the agent's reasoning or behavior.
- Ingestion points: External data sources retrieved via a subagent from URLs, documentation sites, and technical articles (SKILL.md).
- Boundary markers: The skill does not define specific delimiters or "ignore instructions" warnings to isolate external data from the agent's core logic, though it does mandate a structured reporting format for findings.
- Capability inventory: The subagent is explicitly limited to read-only mode, but the parent agent retains the capability to access project files (such as go.mod or lockfiles) and invoke platform-specific agent coordination tools.
- Sanitization: The instructions do not mention any explicit sanitization, filtering, or escaping of external content before it is processed by the agent.
Audit Metadata