The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages agent lifecycles by executing shell commands including git, tmux, and a Python-based utility script located at a local path (~/.claude/skills/codex-tmux/scripts/run.py). These commands are used to orchestrate parallel tasks and the subsequent review process.
[PROMPT_INJECTION]: High Autonomy Instructions: The skill directs the agent to operate without human intervention, explicitly stating there are no approval gates for planning and launching sub-agents. While this is an intended feature for autonomous efficiency, it removes the opportunity for human oversight during critical file modification and commit phases.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface: The automated Codex Review phase processes codebase changes and git diffs which may contain untrusted data. * Ingestion points: Review Agent reads modified files and git diff output (SKILL.md Step 5a). * Boundary markers: The review prompt lacks explicit delimiters or instructions to disregard embedded content within the files being reviewed. * Capability inventory: The Review Agent is granted the ability to fix issues by writing directly to files and executing git commits. * Sanitization: No sanitization, validation, or filtering of the file content is performed before the agent processes it for the review task.

divide-and-conquer