The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user input through the $ARGUMENTS variable and reads existing project files (source code and tests) to generate new tests and implementation code. This ingestion of external data is a standard attack surface for coding assistants, but the risk is mitigated by the skill's specific focus on testing and its reliance on a local test runner for verification.
Ingestion points: $ARGUMENTS in SKILL.md, and project files identified via globbing in Step 0.
Boundary markers: None explicitly defined for untrusted data interpolation.
Capability inventory: File writing (Step 2) and shell command execution (Step 0/3).
Sanitization: Relies on the agent's logic and user oversight for the generated code.
[COMMAND_EXECUTION]: To verify the TDD cycles, the skill identifies and executes local shell commands such as npm test, pytest, or go test based on the project's configuration files (e.g., package.json, pyproject.toml). This execution is a core requirement of the TDD workflow and is confined to the project's defined test environment.

tdd