Skip to content

[Bug]: ACPX Codex worker fails when model/thinking overrides are configured #74305

@SimSef

Description

@SimSef

Bug type

Crash (process/app exits or hangs)

Beta release blocker

No

Summary

Codex ACP worker runs succeed without internal model/thinking overrides, but fail with AcpRuntimeError: Internal error when model and model_reasoning_effort overrides are passed to codex-acp.

Steps to reproduce

  1. Start OpenClaw 2026.4.26 with ACPX enabled and Codex configured as an ACP agent.
  2. Configure the ACPX Codex command to run codex-acp with -c model=gpt-5.5 -c model_reasoning_effort=medium.
  3. From an OpenClaw agent, call sessions_spawn with runtime: "acp", agentId: "codex", mode: "run", and streamTo: "parent".
  4. Ask the ACP worker to perform a web research task.
  5. Observe the spawned ACP task fail with AcpRuntimeError: Internal error.

Expected behavior

The Codex ACP worker should run successfully with supported model/thinking overrides, or return a specific validation error if those overrides are unsupported.

Actual behavior

The ACP worker fails with AcpRuntimeError: Internal error; the ACP stream log only records lifecycle start/error events, and no child transcript file is created.

OpenClaw version

2026.4.26 (be8c246)

Operating system

Linux 6.17.0-1011-azure x64, Node 22.22.2

Install method

npm global: /home/azureuser/.npm-global/bin/openclaw

Model

Override under test: gpt-5.5 with model_reasoning_effort=medium

Provider / routing chain

OpenClaw agent -> ACPX plugin -> @zed-industries/codex-acp@^0.11.1 -> Codex CLI

Additional provider/model setup details

ACPX Codex command used shared Codex auth via CODEX_HOME=/home/azureuser/.codex.

ACPX config included permissionMode: "approve-all" and nonInteractivePermissions: "fail".

Direct Codex CLI sanity checks with -c model=gpt-5.5 -c model_reasoning_effort=medium succeeded on the same VM/auth setup.

Removing the model/thinking overrides from the ACPX Codex command allowed the ACP worker flow to work again.

Logs, screenshots, and evidence

Observed OpenClaw log line:

[ws] ⇄ res ✗ agent errorCode=UNAVAILABLE errorMessage=AcpRuntimeError: Internal error runId=f1b71913-b95d-4dea-a691-97596b7af9f6 error=AcpRuntimeError: Internal error: code=ACP_TURN_FAILED

ACP stream log contained only lifecycle start/error events:

phase: start

phase: error

error: Internal error

No child transcript file was created for the failed ACP session.

Impact and severity

Affected: ACPX Codex worker runs where the operator wants to control the child Codex model/thinking level.

Severity: Medium-high, because ACPX works only when the child model/thinking overrides are omitted.

Frequency: Observed on the tested ACPX Codex spawn attempts with model/thinking overrides configured.

Consequence: Operators cannot reliably configure the internal Codex ACP model/thinking level for spawned ACP workers.

Additional information

The same ACP workflow works after removing -c model=gpt-5.5 -c model_reasoning_effort=medium from the ACPX Codex command.

Metadata

Metadata

Assignees

No one assigned

    Labels

    P2Normal backlog priority with limited blast radius.bugSomething isn't workingbug:crashProcess/app exits unexpectedly or hangsclawsweeper:linked-pr-openClawSweeper found an open linked pull request for this issue.clawsweeper:no-new-fix-prClawSweeper does not recommend queueing a new automated fix PR for this issue.clawsweeper:source-reproClawSweeper found a high-confidence source-level issue reproduction.impact:auth-providerAuth, provider routing, model choice, or SecretRef resolution may break.issue-rating: 🦞 diamond lobsterVery strong issue quality with high-confidence source-level or clear reproduction.

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions