The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read neighboring files and existing tests to generate new unit tests. This creates an ingestion surface for untrusted content if the repository files contain malicious instructions designed to influence the agent's behavior.
Ingestion points: The skill reads implementation files, neighboring *.test.ts files, and the repository's test skill.
Boundary markers: Absent. No instructions are provided to delimit or ignore instructions that might be embedded in the code or tests being read.
Capability inventory: The skill directs the agent to run test files and package-level test scripts using the project's test runner.
Sanitization: Absent.
[COMMAND_EXECUTION]: The skill requires the agent to execute test files and package scripts. While standard for a testing utility, this capability allows for the execution of code contained within the repository's test files.
[NO_CODE]: This skill consists entirely of natural language instructions and configuration metadata; no executable scripts or binaries are included in the skill definition itself.

w-unit