tdd-workflows-tdd-cycle
Pass
Audited by Gen Agent Trust Hub on Nov 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill uses the
$ARGUMENTSvariable, which represents untrusted user input, and interpolates it directly into natural language prompts sent to various subagents (e.g.,architect-review,test-automator,backend-architect). This pattern allows an attacker to inject instructions that could influence the behavior of those subagents. - Ingestion points: The
$ARGUMENTSvariable is interpolated into prompts across multiple phases (Phase 1 through Phase 6) inSKILL.md. - Boundary markers: There are no delimiters (such as XML tags or triple quotes) or explicit instructions to ignore embedded commands used around the
$ARGUMENTSinterpolation. - Capability inventory: The skill leverages the
Tasktool to invoke subagents with significant capabilities, including code generation (backend-architect), test generation and execution (test-automator), and architectural analysis (architect-review). - Sanitization: The instructions do not define any validation, filtering, or escaping mechanisms for the
$ARGUMENTSinput before it is passed to subagents.
Audit Metadata