The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run within lib/integrator.py to perform Git operations such as git add, git commit, and git revert. These calls use list-based arguments rather than shell strings, which effectively prevents shell injection vulnerabilities. Execution is limited to the plugin's root directory and is used for core functionality like version control of changes and rollback support.
[DATA_EXPOSURE & EXFILTRATION]: The skill implements a _validate_path method and _is_path_within checks in lib/integrator.py to ensure all file operations are confined to the target plugin's directory. It specifically checks for and rejects symlinks that point outside the authorized root to prevent directory traversal attacks. File writes use an atomic temp-and-replace strategy to maintain filesystem integrity.
[INDIRECT_PROMPT_INJECTION]: The skill modifies agent instructions by ingesting and inserting protocol templates. It mitigates injection risks through several layers:
Ingestion points: Processes markdown files (.md, .markdown) and plugin.json within the user-specified plugin path.
Boundary markers: Utilizes explicit sentinel markers () to isolate the injected memory protocol from the rest of the plugin's instructions.
Capability inventory: Possesses capabilities for file read/write, directory globbing, and Git command execution via subprocess.
Sanitization: The lib/template_validator.py module scans templates for executable patterns such as shell command substitutions ($() and eval statements before allowing them to be integrated.
[EXTERNAL_DOWNLOADS]: The skill lists ruamel.yaml and PyYAML as optional dependencies in README.md for enhanced YAML processing. These are well-known, standard libraries for configuration management and do not pose a security risk in this context.

integrate