Compaction hits summary output cap with GPT-6 Astra at max reasoning

What happened?

Context compaction fails intermittently with openai/gpt-6-astra at max reasoning. Context-overflow recovery stops with:

Context overflow recovery failed: Summarization failed: generation hit the token cap and the summary is incomplete

Steps to reproduce

1. Launch:

   pi --provider openai --model gpt-6-astra --thinking max
   

2. Run long, tool-heavy conversation until context-overflow recovery triggers. If summary generation exhausts output budget, recovery fails with error above.

Expected behavior

• User does not see above error and auto compaction just works.

Version

@earendil-works/pi-coding-agent 0.85.1; provider/model: openai/gpt-6-astra; reasoning: max.