The Agent Skills Directory

[COMMAND_EXECUTION]: The skill detects and executes local development and validation tools (e.g., npm test, pytest, tsc, cargo test, go vet, ruff, eslint) based on the project's configuration files. These commands are used to establish a baseline and verify that readability improvements do not introduce functional regressions.
[CREDENTIALS_UNSAFE]: In Step 5, the skill performs a grep search across non-code files, including .env, to identify stale references to renamed code symbols. While .env is a sensitive file path typically containing credentials, the access is limited to verifying symbol consistency and results are reported as warnings to the user. No data exfiltration or external network transmission is attempted.
[INDIRECT_PROMPT_INJECTION]: The skill processes source code and configuration files provided as inputs, which represents a surface for indirect prompt injection if those files contain malicious instructions aimed at the cleaning agent.
Ingestion points: Project source files and configuration files identified through $ARGUMENTS (SKILL.md).
Boundary markers: None explicitly defined to separate untrusted file content from agent instructions.
Capability inventory: Reading and writing local files; executing local build, test, and linting tools via shell commands (SKILL.md).
Sanitization: No specific sanitization or escaping of external content is mentioned before processing.

readability-method