The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses a placeholder $ARGUMENTS in its command files to interpolate user input directly into the primary instruction set. This creates a surface for indirect prompt injection where malicious instructions embedded in the user-provided code or arguments could potentially influence the agent's behavior or safety constraints.
Evidence: The files commands/refactor-clean.md and commands/tech-debt.md both utilize the $ARGUMENTS variable without accompanying boundary markers (like XML tags) or specific instructions to disregard any embedded commands within that input.
Capability Inventory: The skill is designed for highly capable environments like claude-code and agent-zero which typically possess file system access and shell execution capabilities.
Sanitization: No explicit sanitization or validation of the $ARGUMENTS content is defined in the provided files.
[EXTERNAL_DOWNLOADS]: The skill's documentation and example configurations reference several well-known third-party security and analysis tools. These are documented neutrally as standard industry integrations.
Evidence: References to tools such as Snyk, CodeQL, SonarQube, and various official GitHub Actions (e.g., github/copilot-autofix, coderabbitai/action) are found in agents/code-reviewer.md and commands/refactor-clean.md.

code-refactoring