The Agent Skills Directory

[COMMAND_EXECUTION]: The launch.py script uses the subprocess module to execute git commands and other Python-based runners. All calls use argument lists and avoid shell execution, which prevents command injection vulnerabilities. These operations are essential to the skill's core purpose of orchestrating development tasks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests markdown files (task contracts and notes) that are used as context for model runners. This potential attack surface is addressed through security measures such as SHA-256 digest validation of all inputs, the use of isolated git worktrees to prevent unauthorized changes to the main repository, and the addition of explicit 'Execution' and 'Review' boundary markers in the generated prompts to constrain model behavior.
[SAFE]: The skill implements significant defensive measures, including validation of all path components using strict regex to prevent traversal, requiring user approval for task routing plans before execution, and verifying the provenance of scripts and models via receipts and checksums. No malicious patterns were detected.

implement-and-review