#Codex 5.3 + OpenClaw keeps asking for approval (even with Ask OFF)

1 messages · Page 1 of 1 (latest)

tardy verge
#

Anyone here using Codex 5.3 with OpenClaw? Need some help 🙏
My agent keeps asking for approval every single time it tries to do anything. It’s honestly super annoying 😅
I already turned Ask = OFF in the UI settings
and set Security = Full
But my agent still requests approval for literally every action.
Is there some agent-level config that overrides the UI settings?
Or is this just how Codex 5.3 behaves right now?
Has anyone figured out how to fully disable the constant approval prompts? 🙏

urban trench
#

I found the same thing - Sonnet = understands agent and soul file which gives it good guidelines for autonomy. Codex = needs hand holding.

scenic heartBOT
#

There are two different “approval” things that often get conflated:

1) OpenClaw “exec approvals” (popups / approval-required for host commands)

This is the gateway/node host safety interlock for exec when the agent tries to run commands on a real machine (host=gateway or host=node). It’s controlled by:

  • Per-session overrides: /exec … and /elevated …
  • Gateway config: tools.exec.ask/security and per-agent agents.list[].tools.exec.*
  • The local approvals store on the execution host: ~/.openclaw/exec-approvals.json (and it’s per-agent inside that file)

Common gotchas when you “set Ask=OFF + Security=Full” but it still prompts:

  • You changed settings for the wrong agent (approvals are per-agent; check the agent selector in the approvals UI).
  • Your gateway config still has tools.exec.ask: "always" (effective policy is the stricter of config + approvals defaults).
  • You’re executing on a node (macOS app / node host), but you changed approvals on the gateway (or vice-versa).

Things to try:

  • In the chat/session with your bot:
    • send /exec to see what it’s currently using