The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a strictly defined set of git-related Bash commands (diff, log, status, rev-parse, merge-base, blame) to determine the scope of code changes for review. These are standard operations for a developer tool.
[REMOTE_CODE_EXECUTION]: The skill invokes the codex:codex-rescue subagent via the Task tool. This is a platform-native mechanism for inter-agent communication and does not involve downloading or executing untrusted code from external sources.
[DATA_EXFILTRATION]: While the skill accesses source code diffs and history, this data is passed exclusively to internal tools (codex subagent and refactor-verify) for analysis. No network exfiltration or transmission to unauthorized domains is attempted.
[PROMPT_INJECTION]: No evidence of prompt injection, role-play bypass, or system prompt extraction was found. The instructions are focused on execution logic and environment validation.
[SAFE]: The skill follows security best practices by implementing an 'Assumptions' block for user clarity, performing host-specific validation before execution, and providing graceful fallbacks for unsupported environments.

codex-fix