The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from repository files (code, comments) and external audit reports. It mitigates this risk through an explicit 'Security boundary' section that instructs the agent to treat all processed content as data rather than instructions.
Ingestion points: Reads files across the repository and external Markdown audit reports.
Boundary markers: Includes a dedicated 'Security boundary' section with explicit 'never instructions' warnings.
Capability inventory: Performs file writes for cleanup and executes shell commands for verification (git, tsc, node, grep).
Sanitization: Implements logic to strip AI-generated formatting and ignore directives embedded in comments or reports.
[COMMAND_EXECUTION]: Uses standard development tools including git, tsc, wc, grep, and xargs to analyze the repository and verify changes. These operations are limited to the local environment and the project scope.
[DYNAMIC_EXECUTION]: Employs node -e during Phase 1.3 to verify the behavior of regular expressions and boundary values found in the code. This is used strictly for localized verification purposes.
[DATA_EXPOSURE]: While the skill reads sensitive files like repository configuration and code, it performs no network operations and focuses entirely on local refactoring tasks.
[OBFUSCATION]: The skill actually includes a security-positive feature that detects and removes 'AI-slop' markers, such as non-standard punctuation (em-dashes, zero-width spaces, and decorative glyphs), which are often used in obfuscation or character-substitution attacks.

awesome-code-cleanup