sdd
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill extracts requirements and evidence from user input and interpolates them into sub-agent prompts for execution, creating a potential vector for indirect prompt injection.\n
- Ingestion points: Objective, requirements, and evidence are extracted in the 'Frame the work' section of SKILL.md and passed as inputs to the 'Agent' tool.\n
- Boundary markers: The instructions do not mandate the use of delimiters (e.g., triple quotes or XML tags) or safety preambles when wrapping untrusted requirements in agent prompts.\n
- Capability inventory: Orchestrated sub-agents (Worker and Reviewer) are intended to perform file-system modifications and run 'verification commands', which are powerful capabilities.\n
- Sanitization: The skill does not prescribe sanitization, escaping, or validation of the requirements text before it is used to generate prompts for sub-agents.\n- [COMMAND_EXECUTION]: The orchestration protocol involves the definition and execution of shell-based verification commands by sub-agents.\n
- Evidence: Sections 1 ('Frame the work') and 3 ('Dispatch ready work') of SKILL.md direct the agent to include 'verification commands' in the task descriptions and sub-agent prompts.
Audit Metadata