The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source files (HTML, JSX, TSX, etc.) as part of its primary workflow. Ingestion points include the detect-slop.cjs script and the audit workflow described in SKILL.md. While this creates an attack surface for indirect prompt injection, the skill includes explicit boundary markers and instructions in both SKILL.md and rules/safe-editing.md to treat all ingested content (comments, UI strings, and tool outputs) as data rather than instructions. Capabilities are restricted to file reading and scoped editing, with a strict safety boundary preventing network access or credential retrieval.
[SAFE]: The skill implements robust safety controls, explicitly prohibiting unauthorized network operations, credential access, and production environment changes. The provided Node.js scripts rely on standard file system modules and do not utilize dangerous functions like eval or exec. All analysis is performed locally on specified target directories.

ai-design-slop-remover