The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs sensitive shell operations including rm for file deletion and git commit for repository state changes. It also executes language-specific linters and build tools (npx, vulture, staticcheck, go vet, mypy) which involve extensive read/write access to the project directory.
[REMOTE_CODE_EXECUTION]: By calling npm test, npm run build, and npx, the skill may execute code defined in local configuration files or download and run packages from external registries. These tools execute in the context of the agent and can be used to run arbitrary scripts if the project configuration is untrusted.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core function of processing untrusted source code.
Ingestion points: Reads files across the src/ directory using grep and static analysis tools.
Boundary markers: None; the skill does not implement delimiters or specific instructions to prevent the LLM from obeying natural language commands hidden within code comments or string literals.
Capability inventory: Includes file deletion (rm), git management (git add, git commit), and arbitrary shell execution through test and build runners.
Sanitization: The skill does not validate or sanitize the content of the files it scans before using that data to determine deletion logic or batching strategies.

dead-code-removal