build-evidence-map
Pass
Audited by Gen Agent Trust Hub on Aug 5, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it instructs the agent to summarize and incorporate excerpts from untrusted external sources and local files.
- Ingestion points: External URLs and local file paths are specified as primary evidence sources in the SKILL.md workflow.
- Boundary markers: There are no instructions for the agent to use specific boundary markers or "ignore" delimiters when reading or extracting content from external source regions.
- Capability inventory: The agent is authorized to execute a bundled Node.js script (scripts/validate.mjs) and utilize the doubt CLI tool.
- Sanitization: No explicit sanitization or filtering logic is prescribed for the excerpts retrieved from external sources before they are added to the .doubt.json artifact.
- [COMMAND_EXECUTION]: The skill requires the agent to run a provided Node.js script (scripts/validate.mjs) for deterministic validation of the evidence map's integrity and graph structure.
- [EXTERNAL_DOWNLOADS]: The skill references an external CLI package (doubt-ai@0.8.0) for rendering and verification tasks. It mitigates risk by explicitly instructing the agent not to install or execute the tool implicitly if it is not already present on the user's system.
Audit Metadata