The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted code snippets through the $ARGUMENTS parameter, creating a vulnerability surface for indirect prompt injection where malicious instructions could be embedded in the code to be refactored.\n
Ingestion points: The $ARGUMENTS variable in the SKILL.md file receives external code input.\n
Boundary markers: The instructions do not define clear delimiters or includes directives for the agent to ignore instructions embedded within the code input.\n
Capability inventory: The skill specifies capabilities for code analysis, modification, and the execution of static analysis tools and unit tests across multiple programming environments.\n
Sanitization: No sanitization or validation of the provided code input is mentioned.\n
Note: The skill contains self-referential statements in the 'Behavior preservation and evidence' section that attempt to define its own security status ('No code from the playbook was executed'), which is a pattern designed to influence security evaluations.\n- [EXTERNAL_DOWNLOADS]: The playbook recommends using the official GitHub CodeQL action (github/codeql-action/analyze@v3) for security and quality analysis. This reference targets a well-known service from a trusted organization within a standard CI/CD workflow context.\n- [COMMAND_EXECUTION]: The instructions and resources mention the use of standard development and static analysis tools such as ruff, mypy, eslint, and semgrep. The execution of these tools is a documented part of a professional development and auditing process.

code-refactoring-refactor-clean