Skip to content

Surface actionable provider startup errors in threads and delegated tasks #29534

Open

Description

@eddietex

What happened

A Codex parent delegated a review to Factory Droid. The child showed no activity for over 20 minutes while appearing to run; follow-up messages to the parent did not start. Reloading did not help; restarting cancelled the review.

A direct Droid thread later failed with an alert containing a generic provider-session error and a long internal identifier. The same generic error remained after dismissing the alert; neither gave authentication guidance.

Diagnosis

The child initialized ACP but never started a run. Session creation failed after ten minutes, authentication failed after another ten minutes, then T3 retried. The reason for the initial stall is unconfirmed.

The direct test server trace reported authentication required and sign-in cancelled, while the thread showed only a generic error. Droid diagnostics reported no usable credentials. Signing in through the CLI restored Droid.

Surface actionable provider failures and recovery steps in direct threads. For delegates, expose startup, authentication and failure states to the parent and user; avoid leaving queued parent messages apparently frozen. Put internal identifiers in diagnostic details.

Steps to reproduce

Observed sequence, not verified as deterministic:

  1. With Droid lacking usable credentials, delegate a review to it from a Codex thread.
  2. Observe the child and send a follow-up to the parent.
  3. Start a direct Droid thread and send a message; inspect the alert and persistent error after dismissal.
  4. Run droid in a terminal, complete sign-in, and retry.

Version

Original freeze: unknown. Direct failure: 0.1.7-nightly.20261102.4080

Environment

macOS arm64, Darwin 25.4.0; desktop and local server; Droid CLI 0.242.1; Codex parent, Factory Droid ACP child and direct thread.

Evidence

Delegated child (UTC):
18:36:26 initialize succeeded; session/new started
18:46:26 session/new failed; authenticate started
18:56:26 authenticate failed; session/new retried
19:01:59 retry interrupted; child had no started run

Direct test trace: AcpRequestError: Authentication required: Sign-in was cancelled. Retry to sign in.

Screenshots:
Image
Image

Related issues

#19612 concerns stale interrupted-child status, rather than startup and authentication failure.

Fix applied or workaround

Run droid in a terminal and complete Factory sign-in. The user confirmed Droid worked afterward.

Filed by

Codex (gpt-6.1-terra) via T3 triage

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    via-triageFiled through npx t3 triage

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions