codex-cli

name	codex-cli
version	1.1.0
description	Run OpenAI Codex CLI for autonomous software engineering tasks via non-interactive `codex exec` with chat-friendly outputs.

Codex CLI Skill

Run OpenAI Codex CLI (codex) from Viber for autonomous coding tasks. This skill uses codex exec (non-interactive), which works reliably from Node/AI SDK tool calls and web UI chat workflows.

Installation

Install Codex CLI globally:

pnpm add -g @openai/codex

Then authenticate:

codex login
codex --version

Tool

• codex_run — Runs Codex CLI with a prompt, returns:
  • summary (status, cwd, mode, exit code)
  • stdoutTail and stderrTail (chat-friendly tail output)
  • output (truncated combined output for deeper follow-up)

Why this is web-UI friendly

• Produces compact response tails so users can quickly see what happened.
• Keeps large command output bounded to protect context windows.
• Returns structured fields (ok, summary, error, exitCode, command) for robust assistant follow-up.
• Validates environment early (Codex installed, cwd exists) with clear user-facing errors.

Parameters

• prompt (required): task for Codex
• cwd (optional): target working directory (must exist)
• waitSeconds (optional): timeout in seconds (default 90, minimum 10)
• approvalMode (optional):
  • full-auto (default): --full-auto
  • auto-edit: writable workspace mode
  • suggest: read-only suggestions
• model (optional): override model (for example gpt-5-codex)

Usage from Viber

Use this skill when user intent is:

• “use codex”
• “run codex on this task”
• “delegate this coding task to codex”

Examples:

codex_run({
  prompt: "Fix failing tests in src/skills/codex-cli.test.ts",
  cwd: "/workspace/openviber",
  approvalMode: "auto-edit",
})

codex_run({
  prompt: "Review this repo and propose a migration plan",
  approvalMode: "suggest",
  waitSeconds: 120,
})

Recommended follow-up behavior in chat

After calling codex_run:

1. Show summary first.
2. Include important lines from stdoutTail / stderrTail.
3. If failed, use error + command to explain next action.
4. Ask user whether to re-run with different approvalMode, model, or cwd.

Testing long or complex Codex tasks

For real-world flows (for example multi-step repo analysis or issue-fix planning), use opt-in live tests:

OPENVIBER_RUN_LIVE_CLI_TESTS=1 pnpm test src/skills/codex-cli.integration.test.ts
OPENVIBER_RUN_LIVE_CLI_COMPLEX_TESTS=1 pnpm test src/skills/codex-cli.integration.test.ts

Notes:

• Complex live tests are suggest mode (read-only) by default for safety.
• They may still fail if codex login has not been completed in the environment.
• Even on failure, return shape remains structured (summary, error, tails) for UI resilience checks.