The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestrator script (review-panel.sh) and its adapters execute shell commands to perform git operations, invoke model CLIs (claude, gemini, codex), and run the Python-based merge engine (merge.py). This is the intended core functionality of the skill.
[EXTERNAL_DOWNLOADS]: The skill interacts with external AI model services (Anthropic, Google) via their official CLI tools. These communications are part of the standard review workflow and target well-known, trusted service providers.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it interpolates external data (git diffs or plan files) into the prompts sent to the AI reviewers in reviewers/claude.sh and reviewers/gemini.sh. However, the risk is minimized as the system only consumes the resulting structured JSON verdicts and does not execute code contained within the diffs.
[DYNAMIC_EXECUTION]: Several scripts use Python heredocs (`python3
<<'PY'`) to execute small snippets of code for JSON parsing and data transformation. This approach is used to maintain portability and avoid cluttering the environment with temporary script files, representing standard developer practice in this context.

review-panel