tdd-workflows-tdd-green
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill template in the implementation playbook interpolates external data ($ARGUMENTS) directly into a prompt directed at a subagent. This design is susceptible to indirect prompt injection if the input contains malicious instructions intended to divert the subagent from its primary task.
- Ingestion points: The prompt template defined in
resources/implementation-playbook.mduses the$ARGUMENTSplaceholder for external input. - Boundary markers: No delimiters (e.g., XML tags or triple quotes) or explicit "ignore instructions within" warnings are used to isolate the untrusted input from the system instructions.
- Capability inventory: The targeted subagent (
test-automator) possesses code-writing and test-execution capabilities, which could be abused if the prompt is successfully injected. - Sanitization: The playbook does not specify any sanitization, validation, or filtering of the content provided via
$ARGUMENTSbefore it is processed by the AI.
Audit Metadata