The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external, untrusted data such as "requirements and user stories" and "code changes" to drive development workflows. This creates a surface for indirect prompt injection, where malicious instructions embedded in these inputs could attempt to override the agent's logic or trigger unintended actions in connected tools.
Ingestion points: The agent processes "requirements and user stories" and "code changes" (referenced in SKILL.md).
Boundary markers: The skill lacks explicit instructions or delimiters to help the agent distinguish between its own core instructions and potentially malicious commands embedded in external requirements.
Capability inventory: The skill mentions integration with high-privilege environments, including build systems (Maven, Gradle, npm, Cargo, MSBuild), IDE plugins, and Continuous Integration (CI) pipelines.
Sanitization: There are no requirements for input validation, sanitization, or human-in-the-loop verification before the agent acts on processed requirements.
[DYNAMIC_EXECUTION]: The skill promotes "Natural language to test code conversion and automation" and "AI-powered test data creation." This behavior involves generating executable code from non-executable descriptions. If the agent automatically saves and executes this generated code in a development or CI environment, it risks executing unintended logic if the generation process is manipulated by indirect injection or model hallucinations.

tdd-orchestrator