v1.22.1

What's Changed

• Add SWE-Bench Pro to run-eval workflow by @neubig in #2959
• perf: bound InMemoryFileStore with LRU eviction by @csmith49 in #3163
• feat(agent-server): base conversation worktrees on origin default branch by @rbren in #3215
• chore(agent-server): drop variable-sized payloads from info logs by @rbren in #3216
• chore(sdk): drop variable-sized payloads from info logs by @rbren in #3217
• docs(codereview): add Logging Hygiene rule against object dumps in info logs by @rbren in #3218
• perf: change Condensation.forgotten_event_ids from list to set by @csmith49 in #3156
• perf: dispatch PubSub notifications concurrently by @csmith49 in #3162
• fix: always close tool executors on conversation close by @csmith49 in #3157
• fix: eliminate TOCTOU race in EventService.send_message by @csmith49 in #3160
• fix(acp): inject system suffix once instead of every turn (#3221) by @simonrosenberg in #3226
• fix(sdk): map typed LiteLLM 401/403 exceptions to LLMAuthenticationError by @VascoSch92 in #3222
• fix(agent-server): ignore redundant WebSocket auth control frames by @VascoSch92 in #3224
• ci(release-binaries): release window binaries by @VascoSch92 in #3228
• fix(agent-server): bundle rich submodules for unicode width data by @VascoSch92 in #3229
• Primary-surface-first extraction ordering to prevent truncation-based bypass by @Fieldnote-Echo in #2709
• feat(agent-server): unify ACP conversation endpoint by @neubig in #3227
• perf(agent-server): make event search O(limit) instead of O(total) by @rbren in #3197
• perf: dedicated thread pool for conversation execution by @csmith49 in #3169
• Fix/laminar/remove litellm callback by @dinmukhamedm in #3008
• fix: cap per-conversation WebSocket subscriber count by @csmith49 in #3166
• perf: batch ConversationState autosaves inside context manager by @csmith49 in #3165
• perf: centralize lease renewal into one background task by @csmith49 in #3168
• test(agent-server): cover persisted settings migrations by @neubig in #3232
• ci: also relock enterprise/poetry.lock in version-bump PR workflow by @aivong-openhands in #3244
• refactor: optimize caching for public skills loading by @hieptl in #3253
• Prepare AgentSettings removal helpers by @neubig in #3208
• Separate configured and effective LLM token limits by @neubig in #3255
• feat: add --extra-python-path / OH_EXTRA_PYTHON_PATH for custom tool imports in binary builds by @xingyaoww in #3240
• examples: standardize directory entrypoints by @xingyaoww in #3259
• fix(llm): stop bleeding LLM credentials through os.environ (#3138) by @csmith49 in #3258
• fix(sdk): unregister atexit handler in LocalConversation.close() by @csmith49 in #3245
• fix: guard litellm modify_params context by @he-yufeng in #3248
• Release v1.22.1 by @all-hands-bot in #3261

New Contributors

• @he-yufeng made their first contribution in #3248

Full Changelog: v1.22.0...v1.22.1