The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from the local project (e.g., package.json, pyproject.toml) and user input ($ARGUMENTS) to determine the test execution environment and specific test cases to run. This presents an indirect prompt injection surface.
Ingestion points: Project configuration files and $ARGUMENTS are used to define the execution environment and test targets.
Boundary markers: The skill lacks specific delimiters or "ignore embedded instructions" warnings when interpolating these values into tool prompts or shell commands.
Capability inventory: The skill has the capability to execute shell commands for various test runners (npm, pytest, go test, etc.) and performs broad directory discovery via file system globs.
Sanitization: No instructions are provided to sanitize or validate strings retrieved from the codebase or $ARGUMENTS before they are passed to shell execution tools.
[COMMAND_EXECUTION]: The agent is instructed to identify and run test commands (stored as TEST_CMD) found in project files. A malicious actor could provide a project containing a crafted package.json or Makefile designed to execute arbitrary code when the agent attempts to run the project's tests.

tdd