worktree-spinoff

Pass

Audited by Gen Agent Trust Hub on Nov 7, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The orca-spinoff bash script utilizes embedded Python logic (defined in the PYLIB variable and the pa/py helper functions) to handle JSON processing, URL parsing, and complex control flow. These scripts are executed dynamically at runtime using python3 -I -c.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests external data through the --handoff and --prompt-file parameters. This data is interpolated into the system prompt for a new agent terminal. While the script uses shell-level escaping to prevent command injection during the launch process, it does not filter or sanitize the file content for adversarial instructions directed at the agent.\n
  • Ingestion points: Files provided via --handoff or --prompt-file in scripts/orca-spinoff.\n
  • Boundary markers: The prompt template uses newline separators and a standard instruction header to delimit the handoff content.\n
  • Capability inventory: The script performs git operations (push, branch, worktree), filesystem modifications (symlinking via ln -s, directory creation, file copying), and agent terminal management via the orca CLI.\n
  • Sanitization: Employs sed-based escaping for the single-quote character to prevent shell injection into the agent launch command.\n- [COMMAND_EXECUTION]: The script automates complex repository and workspace management tasks by executing various system and development commands, including git worktree and git push. It also facilitates the symlinking of untracked directories (such as .env or node_modules) into the new worktree based on user-supplied paths.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 7, 2026, 02:13 AM