v0.8.55
Summary
- OpenAI tool_calls pairing: per-
tool_use_idsanitize removes partially satisfied assistant tool calls so the API no longer returns 400 for orphanedtool_call_ids. - Agent loop cancel: synthetic error
ToolResults for unexecuted tools; pairing sanitize after each tool round and before every LLM request.
Test plan
- Parallel tool call → cancel mid-batch → next user message succeeds without OpenAI 400
-
cargo test sanitize_tool_useinpiscis-kernel
Full Changelog: v0.8.54...v0.8.55