The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided behavior descriptions to generate and execute code. This creates a surface where malicious instructions could be embedded in the requirements.
Ingestion points: The [behavior or slice] input parameter in SKILL.md serves as the source of instruction for the cycle.
Capability inventory: The skill implements code changes and executes tests using system tools.
Boundary markers: The skill does not explicitly mention sanitizing user input for safety, though it emphasizes tracing flows and reusing existing code, which restricts the scope of new implementation.
Sanitization: None explicitly described; however, the structured TDD process (red-green-refactor) involves verification steps that act as natural review points.
[DYNAMIC_EXECUTION]: The skill is designed to write and run code dynamically as part of its TDD workflow.
Implementation: The agent writes production code and tests based on input.
Execution: The workflow requires running tests to verify failures and successes.
Context: This execution is the intended primary purpose of the skill and follows standard development practices, including advice on mocking external boundaries rather than internal behavior.

hl-tdd