Codex Agent

Interact with the Codex agent. This skill wraps input in strict JSON enforcement prompts and handles retries.

Activation

Triggers: "ask codex", "codex help", "consult codex"

Features

• Strict JSON Output: The agent is prompted to return ONLY valid JSON.
• Retry Logic: Automatically retries with stricter prompts if the first response is not valid JSON.
• Schema Validation: Supports passing a JSON schema to validate the output against (if supported by the underlying implementation).

Usage

Simple Prompt

# Usage: .claude/skills/codex/codex-agent.sh "Your prompt here"
.claude/skills/codex/codex-agent.sh "Summarize the history of AI."

JSON Request (Advanced)

You can pass a JSON object with prompt, context, and schema.

{
  "prompt": "Generate a user profile",
  "schema": {
    "type": "object",
    "properties": {
      "username": {"type": "string"},
      "age": {"type": "integer"}
    }
  }
}

.claude/skills/codex/codex-agent.sh '{"prompt": "...", "schema": {...}}'

Response Format

Success:

{
  "key": "value",
  ...
}

Error:

{
  "ok": false,
  "error": {
    "code": "error_code",
    "message": "Description"
  }
}