00-plan-assembly
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides detailed instructions in
references/plan-figures.mdfor executing shell commands and Python modules, such asbash scripts/build-doc.shandpython -m engine, to validate diagrams and build the final document package using a local 'SRS engine'. These operations are presented as the standard method for figure rendering and document construction. - [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied business plan content, creating a risk of indirect prompt injection.
- Ingestion points: Untrusted data enters the agent context through 'Completed or near-complete plan sections', 'financials', and 'evidence registers' as defined in
SKILL.md. - Boundary markers: The skill lacks explicit delimiter-based sanitization or instructions to ignore embedded commands within the processed sections, though it does mention reconciling conclusions to sources.
- Capability inventory: The agent is instructed to execute shell scripts and Python tools (
bash scripts/build-doc.sh,python -m engine) to process and render the final document. - Sanitization: No evidence of input validation, escaping, or filtering for instructions hidden in the plan sections is provided in the documentation.
Audit Metadata