mirror of
https://github.com/AstrBotDevs/AstrBot.git
synced 2026-11-22 22:23:18 +08:00
* stage * refactor: linearize conversation events and reclaim plugin snapshots * test: adapt master tests to event-context APIs after merge - test_tool_loop_agent_runner: restore @pytest.mark.asyncio on test_step_notices_reach_model_once_per_threshold (lost in merge). - test_cron_manager: give the build_main_agent / runner mocks the conversation_events field and an awaitable aclose, matching the event-writer wiring in cron/manager.py. * style: apply ruff formatting to exclude_ids filter in sqlite.py * test: adapt gemini/message-tools tests to event-context APIs - test_gemini_source: fake_retry now accepts request_event_recorder, matching retry_provider_request's new keyword argument. - test_message_tools: _DummyRespondEvent exposes conversation_events, which call_event_hook now reads when emitting the after-send hook. * fix: merge duplicate chat history translations * refactor: consolidate plugin conversation APIs * refactor: bind conversation writers to agent execution context * refactor: separate conversation migration and display projection Remove deprecated conversation metadata methods and align architecture and plugin documentation with the current APIs. * refactor: make conversation writer ownership explicit Use per-run hooks instead of ambient writer and plugin identity bindings. Require explicit conversation IDs for append APIs and separate persistence from execution context mutation. * fix: harden conversation event writes and request capture * refactor: remove legacy checkpoints and unify conversation history * perf: incrementally project acknowledged conversation events * refactor: clarify agent event and persistence boundaries * docs: remove event context design notes * perf: reduce conversation SQL round trips and support PostgreSQL * refactor: remove PostgreSQL support while retaining SQL optimizations * refactor: store raw context once and filter temporary content on replay * test: retain essential event context regression coverage * refactor: derive conversation revisions from events Remove persisted event head, context leaf and replay baseline pointers. Derive positions through indexed queries, migrate existing databases and preserve backup compatibility. Refresh conversation activity at turn settlement instead of every execution event. * refactor: use compact IDs for conversation events