Auto capute not working through claude code

[ehsanghorbani190]

Hi! im using opencode-claude-auth to use claude code inside opencode. I can see that anthropic is available as a provider and the model i've specified in the config, is present on the provider models list. but the auto capture just fails, because of timeout.
This is the logs im seeing:

timestamp=2026-08-26T12:55:56.335Z level=INFO run=003ecb68 message="llm runtime selected" llm.runtime=ai-sdk llm.provider=anthropic llm.model=claude-haiku-4-5-20251001
timestamp=2026-08-26T12:55:58.292Z level=INFO run=003ecb68 message=loop session.id=ses_fc1dc0c70ffef3toT4YqKmre6I step=7
timestamp=2026-08-26T12:55:58.306Z level=INFO run=003ecb68 message=process session.id=ses_fc1dc0c70ffef3toT4YqKmre6I messageID=msg_03e244056001snJYj2ipYCCM2c
timestamp=2026-08-26T12:55:58.308Z level=INFO run=003ecb68 message=stream providerID=anthropic modelID=claude-haiku-4-5-20251001 session.id=ses_fc1dc0c70ffef3toT4YqKmre6I small=false agent=opencode-mem-structured mode=subagent
timestamp=2026-08-26T12:55:58.309Z level=INFO run=003ecb68 message="llm runtime selected" llm.runtime=ai-sdk llm.provider=anthropic llm.model=claude-haiku-4-5-20251001
timestamp=2026-08-26T12:56:00.368Z level=INFO run=003ecb68 message=loop session.id=ses_fc1dc0c70ffef3toT4YqKmre6I step=8
timestamp=2026-08-26T12:56:00.420Z level=INFO run=003ecb68 message=process session.id=ses_fc1dc0c70ffef3toT4YqKmre6I messageID=msg_03e244874001HIxipT77w26i3c
timestamp=2026-08-26T12:56:00.421Z level=INFO run=003ecb68 message=stream providerID=anthropic modelID=claude-haiku-4-5-20251001 session.id=ses_fc1dc0c70ffef3toT4YqKmre6I small=false agent=opencode-mem-structured mode=subagent
timestamp=2026-08-26T12:56:00.443Z level=INFO run=003ecb68 message="llm runtime selected" llm.runtime=ai-sdk llm.provider=anthropic llm.model=claude-haiku-4-5-20251001

[lindixu6-hash]

Thanks for the report. The excerpt confirms that the internal opencode-mem-structured session is being selected, but it does not include the final timeout/error or enough version data to distinguish an opencode-mem timeout from an OpenCode / opencode-claude-auth structured-output loop.

One notable signal is that the log reaches step=7 and step=8, while opencode-mem registers this internal agent with steps: 2 and maxSteps: 2. That suggests the host/plugin path may not be honoring the agent step cap, but we need the exact versions and terminal error before assigning the fault boundary.

Could you provide:

1. opencode-mem, OpenCode, and opencode-claude-auth versions;
2. the complete final Auto Capture Failed message and approximate elapsed time (opencode-mem's current hard ceiling is 90 seconds);
3. whether an ordinary chat with the same anthropic/claude-haiku-4-5-20251001 model succeeds;
4. your opencode-mem provider config (opencodeProvider / opencodeModel, with secrets removed);
5. a redacted Claude-auth diagnostic excerpt around the failed internal session. Enable it with CLAUDE_AUTH_DEBUG=1, restart OpenCode, reproduce once, then inspect ~/.local/share/opencode/claude-auth-debug.log.

As a quick isolation check, please also try one provider/model known to support OpenCode format: json_schema. If that succeeds while the Claude-auth model loops, the failure is in the provider compatibility path rather than capture scheduling/storage.