Codex — Orchestrator

Main entry point for all Codex skills. Supports slash commands AND natural language.

Trigger Patterns

Slash command

/codex [subcommand] <prompt|file_path>

Natural language (examples)

• "ask codex to review src/auth.ts"
• "have codex explain the parser"
• "let codex write tests for this"
• "use codex to fix the login bug"
• "run codex on this file"
• "send this to codex"
• "codex should refactor the db module"

Routing Rules

Detect the intent from either the subcommand or natural language:

Intent detected	Route To	Trigger examples
Review / critique / check	codex-review	"review", "check this code", "find bugs in", "audit"
Explain / understand	codex-explain	"explain", "how does this work", "walk me through", "what does this do"
Test / spec	codex-test	"test", "write tests", "add specs", "cover with tests"
Fix / debug / repair	codex-fix	"fix", "debug", "repair", "solve", "the bug in"
(anything else)	Direct exec	Any other task → freeform Codex prompt

Workflow

Step 1: Detect Intent

From $ARGUMENTS or the user's natural language, identify:

1. The intent — review, explain, test, fix, or freeform
2. The target — file path, code reference, or task description

Step 2: Route

• If intent matches a sub-skill: invoke it using the Skill tool, passing the target as args
• If freeform: run Codex directly:

  npx @openai/codex exec --full-auto "<prompt>"
  

  If it fails due to missing git repo, retry with --skip-git-repo-check.

Step 3: Return Output

Return the result to the user.