The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code comments which could contain malicious instructions. \n
Ingestion points: The agent reads source files within the repository to identify and classify comments as keepers or noise. \n
Boundary markers: The instructions do not define specific delimiters for separating the instructions from the data being analyzed. \n
Capability inventory: The agent has the ability to write to the filesystem and execute shell commands (linting, testing, and a verification script). \n
Sanitization: The verify-comments-only.sh script acts as a strong logical control by using esbuild to prove that only non-functional changes (comments/whitespace) occurred.\n- [EXTERNAL_DOWNLOADS]: The verify-comments-only.sh script utilizes package runners like npx, bunx, or pnpm dlx to execute esbuild. These tools may download the esbuild binary from public registries if it is not already cached locally.\n- [COMMAND_EXECUTION]: The workflow involves executing various local commands, including project test suites, linters, git grep for directive auditing, and the provided verify-comments-only.sh script for diff verification.

nuke-comments