Two OpenClaw Settings That Fix 'Lazy' ChatGPT

Fresh OpenClaw

Why GPT feels lazy, the settings to change first, and the before-and-after test workflow.

OpenClaw 4.14 plus 3 config fixes made GPT-5.4 actually usable: a tighter agent contract, the right plugin path, and the right default model.

---

The main issue was that the agent felt weak, slow, dumb, and not proactive. Same model, but the surrounding system was not closing the loop cleanly. These fixes improved the setup.

It was not just a model problem

The real issue was that the agent felt weak, slow, dumb, and not proactive. It was not closing loops cleanly. The fix was to tighten the agent contract, enable the right plugin path, and point the default model at GPT-5.4.

Before (feels like a weak agent):

execution contract: too loose
plugin path:         not helping enough
model behavior:      weak loop closure
agent experience:    feels slow, dumb, and not proactive

After (same model, better system):

openclaw version: 4.14
strict mode:      strict-agentic
plugin path:      codex enabled
agent experience: cleaner routing, recovery, and better loop closure

The three fixes

These are the three lines that were applied:

1. Strict agent contract
   agents.defaults.embeddedPi.executionContract = "strict-agentic"
   (the line that made the agent stricter)

2. Codex plugin enabled
   plugins.entries.codex.enabled = true
   (the plugin path we kept)

3. GPT-5.4 as the default model
   agents.defaults.model = "codex/gpt-5.4"
   (local resolved name may render as openai-codex/gpt-5.4)

The workflow to reproduce it

Back up your OpenClaw system first, update to 4.14, run the master prompt, then verify the three lines by asking the agent if they are enabled.

1. Back up OpenClaw first - back up your openclaw.json config before changing anything.
2. Update to 4.14 - bring OpenClaw to 4.14 first so the workflow matches.
3. Run the master prompt - use it to inspect the config, compare exact values, and apply only the minimum changes required.
4. Ask if the three lines are enabled - verify directly by asking your agent "Are the following enabled?" and checking the three exact config lines.

The master prompt

Optimized around the exact point: update to OpenClaw 4.14, then verify the three specific config lines that made the system stricter and more usable. Make backups before you do anything.

Before we get started, the first thing you always want to do is make a backup. Do that now.

You are verifying the exact OpenClaw setup from this video.

Goal:
Update OpenClaw to 4.14, then verify that the system matches these three intended config lines:
- agents.defaults.embeddedPi.executionContract = "strict-agentic"
- plugins.entries.codex.enabled = true
- agents.defaults.model = "codex/gpt-5.4"

Tasks:
1. Back up the current OpenClaw config before changing anything.
2. Check the installed OpenClaw version.
3. Update OpenClaw to 4.14 if needed.
4. Inspect the live config and verify whether these three lines are present and active:
   - agents.defaults.embeddedPi.executionContract = "strict-agentic"
   - plugins.entries.codex.enabled = true
   - agents.defaults.model = "codex/gpt-5.4"
5. If any of the three lines are missing or different, show the exact current value and the exact intended value.
6. Apply the minimum changes needed to match the intended setup.
7. Re-check the config after changes.
8. Confirm whether the final setup now matches all three lines.
9. Report anything that may render differently in practice, for example if the local resolved model name appears as openai-codex/gpt-5.4.
10. Do not enable extra settings beyond what is required for these three lines unless they are necessary to keep the system working.

Constraints:
- Do not hallucinate success.
- Show exact values before and after.
- Keep the changes minimal.
- If something cannot be verified, say so clearly.

Output format:
- Backup created
- OpenClaw version before update
- Update result
- Verification of line 1
- Verification of line 2
- Verification of line 3
- Exact changes applied
- Final match status
- Rollback notes

Why this makes GPT-5.4 feel better

• Cleaner routing and recovery - 4.14 improves the surrounding runtime behavior, which matters when agents or sub-agents were previously freezing or getting lost mid-task.
• Reliable browser operations - when the surrounding system is configured correctly, the agent stops feeling blocked and can execute with more confidence and follow-through.
• More useful GPT-5.4 work - when the system around the model stops fighting it, GPT-5.4 can execute, verify, and follow through instead of feeling blocked or half-alive.
• Respectable operator flow - the master prompt gives a concrete, auditable workflow instead of another vague "use this model" recommendation.