cawplan-risk-assessment
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data such as ticket titles, descriptions, and feedback. This creates a surface for indirect prompt injection. Evidence: 1. Ingestion points: Data is ingested via 'tickets search' and 'qa-insights risk-assessment compute' commands. 2. Boundary markers: The skill includes 'UNTRUSTED_CONTENT.md' which instructs the agent to ignore instructions in data and use quoting. 3. Capability inventory: The skill can read system data and write assessments to the vendor's database. 4. Sanitization: The workflow enforces user confirmation, dry-run checks, and explicit confirmation flags for all write operations.
- [COMMAND_EXECUTION]: The skill utilizes the Bash tool to run the 'cawplan' CLI for risk computation and management. These commands are executed within a controlled workflow that validates product and version identifiers.
Audit Metadata