copilot-driver --resume fails with 'No authentication information found' after transient AI model error

[tadelesh]

Summary

When the Copilot agent crashes due to a transient AI model server error (5 retries exhausted), all --resume retry attempts immediately fail with "No authentication information found". The authentication token is not properly passed through to the resumed copilot process, making the retry mechanism ineffective.

Environment

• gh-aw version: v0.68.1
• Engine: copilot with claude-opus-4.6
• Workflow: doc-update (workflow_dispatch trigger)
• Repository: Azure/typespec-azure
• Run URL: https://github.com/Azure/typespec-azure/actions/runs/24332968555/job/71043216899

Steps to Reproduce

1. Run an agentic workflow using engine.id: copilot with engine.model: claude-opus-4.6
2. The agent runs successfully for ~39 minutes, producing code changes
3. A sub-agent call fails with a transient AI model server error:

   Failed to get response from the AI model; retried 5 times (total retry wait time: 6.59s)
   Last error: Unknown error
   

4. The copilot process exits with code 1
5. copilot-driver detects partial execution and attempts --resume

Expected Behavior

The --resume attempts should inherit the authentication context from the original run and continue execution from where it left off.

Actual Behavior

All 3 --resume retry attempts fail immediately (within ~2 seconds each) with:

Error: No authentication information found.
Copilot can be authenticated with GitHub using an OAuth Token or a Fine-Grained Personal Access Token.
To authenticate, you can use any of the following methods:
  - Start 'copilot' and run the '/login' command
  - Set the COPILOT_GITHUB_TOKEN, GH_TOKEN, or GITHUB_TOKEN environment variable
  - Run 'gh auth login' to authenticate with the GitHub CLI

Relevant Logs

[copilot-driver] attempt 1: process closed exitCode=1 duration=39m 14s stdout=18542B stderr=499B hasOutput=true
[copilot-driver] attempt 1 failed: exitCode=1 isCAPIError400=false hasOutput=true retriesRemaining=3
[copilot-driver] attempt 1: partial execution — will retry with --resume (attempt 2/4)

[copilot-driver] attempt 2: spawning: /usr/local/bin/copilot --add-dir /tmp/gh-aw/ ... --resume
[copilot-driver] attempt 2: process closed exitCode=1 duration=1s stdout=0B stderr=404B
[copilot-driver] attempt 2 failed: exitCode=1

[copilot-driver] attempt 3: spawning: /usr/local/bin/copilot ... --resume
[copilot-driver] attempt 3: process closed exitCode=1 duration=1s stdout=0B stderr=404B

[copilot-driver] attempt 4: spawning: /usr/local/bin/copilot ... --resume
[copilot-driver] attempt 4: process closed exitCode=1 duration=1s stdout=0B stderr=404B

[copilot-driver] all 3 retries exhausted — giving up (exitCode=1)

Impact

• The --resume retry mechanism is completely broken when auth tokens are not propagated
• Long-running workflows (~39 min) lose all progress and must restart from scratch
• The original run had already produced +365 lines of changes that were lost

Possible Root Cause

The --resume spawn appears to not inherit the GITHUB_TOKEN / COPILOT_GITHUB_TOKEN / GH_TOKEN environment variables that were available during the initial attempt. The auth env var may be cleaned up or expired between attempt 1 exit and the resume spawns.