[Bug] Fable 1M context indicator changes to 200K after session recovery

[Vizards]

Affected area

Agent runtime / ACP

Installation method

Desktop release

Lody version or commit

0.90.1; current main is also affected

Operating system

macOS 15.6 arm64

Agent or runtime

Built-in Claude Code 2.1.258, Fable (claude-fable-5[1m]), Max effort

What happened?

While using Fable Max, the context indicator initially showed 40K / 1M. After the Claude ACP process restarted and Lody restored the session, it changed to 237.9K / 200K and 100% used.

The model was still Fable. The restored turn reported 237,882 tokens and completed successfully, so the 237.9K usage is real but the 200K limit appears to be a display fallback. This makes it look as if the model was downgraded or had exceeded its context window when neither happened.

What did you expect?

Fable should continue to show a 1M context limit. If Lody temporarily does not know the limit during session recovery, it should hide the number until the real value is available instead of showing 200K.

How can we reproduce it?

1. Start a built-in Claude Code session in Lody 0.90.1.
2. Select Fable and Max effort, then confirm that the indicator shows used / 1M.
3. Continue the same session after its Claude ACP process has restarted and the session has been restored.
4. Send another turn and check the context indicator. It may change to used / 200K even though Fable is still selected.

How often does it happen?

Sometimes

Relevant log output

selectedModel=claude-fable-5[1m] effort=max
runtimeModel=claude-fable-5
usageTotal=237882
display="237.9K / 200K"

Additional context

This looks like a regression of #308, which already fixed this behavior by hiding context usage while the limit is unknown.

#308 pointed the Claude adapter submodule to the fixed commit 0164d984. The next Lody commit, bf979598, pointed it back to the older d395b3dc, which defaults an unknown limit to 200K. Current main still uses that older commit.

The adapter fix belongs in LodyAI/acp-extension-claude, but this issue is filed in LodyAI/Lody because that is where the regression was integrated and shipped. To prevent another rollback, the fix should first land on the adapter's default branch and then be referenced by Lody.

Before session recovery (40K / 1M):

After session recovery (237.9K / 200K):

Before submitting

• I searched the existing issues and did not find a duplicate.
• This report concerns an open-source component in this repository, not a hosted service, Web or mobile app, account, or billing issue.
• This is not a security vulnerability; security reports follow the repository's security policy.
• I removed credentials, private source, conversations, prompts, personal data, and other sensitive information.