The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute standard development commands such as git diff, grep, and various build/test tools (npm, pnpm, pytest, cargo) to identify code changes and validate that deletions do not break existing functionality.
[DATA_EXPOSURE]: Technical references in the skill demonstrate high awareness of sensitive data boundaries (auth/authz, secrets, PII) and instruct the agent to treat these as critical security surfaces that require specialized treatment, rather than performing unauthorized access or exfiltration.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface as it processes untrusted code diffs and has the capability to modify or delete code. However, it implements a robust evidence chain and safety protocol:
Ingestion points: Reads from git diff and file system content (SKILL.md).
Boundary markers: Explicitly mandates that validation exists at trust boundaries and protects mandated process workflows (SKILL.md Rules 1 & 2).
Capability inventory: Uses grep for reference checking and standard file-write operations to perform code 'sweeps' (SKILL.md).
Sanitization: Strictly requires running the project's 'narrowest matching gate' (tests or linters) after any modification to ensure behavior remains unchanged.

nuke-lean