closeout-commissioning
Pass
Audited by Gen Agent Trust Hub on May 12, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions in
SKILL.mdspecify automated integration with multiple external data sources to populate checklists and status reports. This creates an attack surface where untrusted data could contain instructions that manipulate the agent's logic or outputs. - Ingestion points: Files
quality-data.json,directory.json,specs-quality.json,project-config.json,procurement-log.json,plans-spatial.json, andchange-order-log.jsonare read and processed during the/closeoutinvocation. - Boundary markers: The instructions do not define clear delimiters or "ignore instructions" wrappers for the data being interpolated into the prompt from these external files.
- Capability inventory: The skill performs complex cross-referencing and status flagging (e.g., verifying O&M manual completeness, checking warranty deadlines, and validating test results against specifications).
- Sanitization: There is no mention of data validation, escaping, or filtering for the external JSON content before it is processed by the agent.
Audit Metadata