The Agent Skills Directory

PROMPT_INJECTION (LOW): Indirect Prompt Injection Surface. The skill includes multiple prompts designed to ingest and act upon externally generated "audit findings."
Ingestion points: Files SKILL.md, CODE.md, DATABASE.md, and DEPENDENCIES.md all contain prompts utilizing the [paste audit findings] pattern.
Boundary markers: Absent. The prompts lack delimiters (such as triple quotes or XML tags) or system instructions to ignore embedded commands within the pasted findings.
Capability inventory: The skill explicitly grants the AI the capability to delete source files (CODE.md), execute destructive SQL commands like DROP TABLE and DROP COLUMN (DATABASE.md), and modify project dependencies via npm uninstall (DEPENDENCIES.md).
Sanitization: No sanitization or validation logic is defined to ensure the pasted findings do not contain malicious instructions aimed at the high-privilege operations the skill supports.
COMMAND_EXECUTION (SAFE): The skill documentation and prompts recommend the use of standard, well-known development and testing tools.
Evidence: Usage of npm audit, npm outdated, depcheck, artillery, and webpack-bundle-analyzer is appropriate for the skill's stated purpose of performance and dependency optimization.

optimize