The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes dynamic context injection (using the !command syntax) in the SKILL.md frontmatter to automatically discover the project's directory structure and configuration files upon loading. These commands (e.g., find) are benign and used solely for environment discovery.
[EXTERNAL_DOWNLOADS]: To ensure compliance with modern standards, the skill fetches action version information from the official release pages of various GitHub and Docker organizations. This is a standard administrative task for CI/CD maintenance.
[REMOTE_CODE_EXECUTION]: The skill provides a template for an 'Auto-fix' workflow that incorporates the anthropics/claude-code-action. This autonomous agent pattern is designed to resolve CI failures and is sourced from a trusted organization.
[PROMPT_INJECTION]: The provided 'Claude Auto-fix' workflow template defines an indirect prompt injection surface because it processes CI failure logs (untrusted external data) and passes them to an LLM-driven action with repository write permissions. The template includes safety mitigations such as loop prevention and instructions for manual review.

configure-workflows