The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/harness.py script executes the git command to identify repository roots. This implementation utilizes list-based arguments and scrubs environment variables to ensure the execution is confined to repository discovery and does not expose the process to command injection or identity hijacking.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external Markdown files for agent handoffs and knowledge management. To mitigate risks associated with untrusted content, the script enforces strict path validation that rejects symbolic links and directory traversal (..), ensuring the agent context is only populated from authorized directories. Ingestion occurs via the status and read operations, while capabilities are restricted to managed file writes and resource claiming within the workspace.

harness-task