codex-skill_skill

Overview

This skill leverages OpenAI Codex/GPT models to autonomously implement code changes, run tests, and report results using the Codex CLI. It supports read-only analysis, workspace-write edits, and an opt-in danger-full-access mode for broader system actions. Use simple codex exec commands or the --full-auto alias to run end-to-end development tasks with minimal supervision.

How this skill works

The skill runs inside codex exec, parsing user requests into a focused execution plan, executing commands, and streaming progress. It uses sandbox modes to control permissions: read-only for analysis, workspace-write for file edits and tests, and danger-full-access for network and system operations. Results are summarized in a structured final report listing changes, metrics, verification steps, and next steps.

When to use it

• Count lines of code, analyze architecture, or produce read-only reports
• Implement features, fix bugs, or refactor code with file edits enabled
• Run tests, CI-like checks, and verify changes automatically
• Perform batch updates across a codebase (imports, APIs, patterns)
• Install dependencies or perform system-level setup when explicitly allowed

Best practices

• Start in read-only mode to gather context before enabling writes
• Use --full-auto (workspace-write) for typical coding tasks to allow edits
• Only enable danger-full-access when running in an isolated environment
• Make reasonable assumptions for minor ambiguities; ask only for critical missing data
• Run relevant tests after changes and include verification in the final report

Example use cases

• codex exec -s read-only "count lines of code by language" to audit a repo
• codex exec --full-auto "fix the login bug" to locate, patch, test, and commit a fix
• codex exec -m gpt-5.2 -s workspace-write "implement user authentication feature"
• codex exec -s danger-full-access "install payment SDK and integrate API" in a container
• codex exec --json -s read-only "scan for security issues and output JSON"