OpenAI Codex Sub Agents

AdvisoryAudited by Static analysis on Apr 30, 2026.

Overview

No suspicious patterns detected.

Findings (0)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

Codex may modify project files and run commands in the selected workspace as part of a coding task.

Why it was flagged

The skill documents a workflow where Clawdbot invokes Codex CLI with `--full-auto`, allowing automated workspace edits and command execution for coding tasks.

Skill content
exec codex exec --full-auto --cd ~/projects/medreport "fix TypeScript errors in src/components"
Recommendation

Use this only on trusted repositories, set `--cd` to a specific project, and prefer read-only or approval-based modes for unfamiliar code.

What this means

Installing or configuring this integration can give Clawdbot access to your Codex/OpenAI-authenticated account context.

Why it was flagged

The integration describes copying Codex OAuth tokens from the local Codex auth file into Clawdbot auth profiles.

Skill content
Clawdbot auto-syncs OAuth tokens from Codex CLI:
- Source: `~/.codex/auth.json`
- Target: `~/.clawdbot/agents/<agentId>/agent/auth-profiles.json`
Recommendation

Only use this with an account and agent configuration you trust; review token storage locations and log out or rotate credentials if access is no longer needed.

What this means

The local behavior depends on the externally installed Codex CLI package and its version.

Why it was flagged

The skill is instruction-only and asks the user to install an external global npm package rather than providing a pinned managed install artifact.

Skill content
npm i -g @openai/codex
Recommendation

Install Codex CLI from the official source, keep it updated intentionally, and consider pinning or reviewing the package version in controlled environments.

What this means

Tasks, code context, and results may move between the main agent and the coding subagent.

Why it was flagged

The integration pattern enables a main agent to spawn or communicate with a Codex-backed coding subagent.

Skill content
tools: {
    agentToAgent: {
      enabled: true,
      allow: ["main", "coder"]
    }
  }
Recommendation

Restrict allowed agents and tools to the minimum needed, and avoid sending sensitive repository or credential data to subagents unnecessarily.

What this means

Prior prompts, task context, or code-related session state may remain on disk after a task is finished.

Why it was flagged

The Codex CLI keeps session files that can persist coding context and be resumed later.

Skill content
Session Files

Location: `~/.codex/sessions/`
Recommendation

Avoid putting secrets in prompts, and periodically review or clear Codex session files if persistence is not desired.