The Agent Skills Directory

Controlled Local Script Execution: The skill invokes a validation script, validate_tracking_source.py, to verify input data integrity. This execution is limited to a specific local path within the plugin's environment.
Shell Injection Protections: The instructions provide explicit guidance to avoid eval and use shell-quoting for all command-line parameters. This practice prevents the execution of arbitrary commands when the agent interacts with the GitHub CLI.
Secure Data Handling for CLI Operations: The skill utilizes temporary files with restricted permissions (0600) to store payloads for CLI writes, ensuring that sensitive finding information is protected from unauthorized access during the tracking process.
Human-in-the-Loop Review: A mandatory preview and approval step is required for all write operations. This ensures that the user can verify the destination, audience, and content before any data is transmitted to external platforms.
Indirect Prompt Injection Mitigations: The skill is designed to ingest data from security scan reports which may contain untrusted content. It mitigates this risk by instructing the agent to treat all input strings strictly as data and by providing a robust inventory of capabilities (file writes, CLI execution) that are protected by sanitization and user review.

track-findings