Skip to content

v0.11.3

Choose a tag to compare

View all tags
@huangzesen huangzesen released this 10 Jun 02:04
v0.11.3
6ff15dd

Highlights

  • Cleaner notification synchronization: ordinary notification sync now injects a synthesized system(action="notification") tool-call/result pair instead of visible diary-like text, keeping conversation history cleaner.
  • Traceability for tool batches: LLM calls now carry an api_call_id through responses and tool lifecycle events, giving the TUI a reliable source for grouping tool batches by model API call.
  • More robust runtime operations: this window includes preset/auth validity guards, MCP restart fixes, daemon configuration improvements, post-molt continuation work, and release/procedure documentation updates.
  • Sharper manuals and memory guidance: bash, psyche, knowledge, and related manuals now better document async agent CLI usage, session-journal sub-knowledge, and release/process expectations.

Runtime and developer changes

  • LLMResponse.api_call_id is logged on llm_call / llm_response and propagated into tool call/result events for downstream renderers.
  • Notification sync now emits tool-only synthesized history, with summary text kept log-only.
  • Knowledge and psyche manuals clarify the routing-parent / child sub-knowledge pattern for session journals.
  • Bash manual explicitly forbids synchronous long-running agent/coding CLI calls and directs agents to async + poll.
  • Preset/auth, MCP, daemon, and post-molt continuation fixes are included from the v0.11.2..v0.11.3 window.

Validation

Release candidate before version bump: 93fb748f8e66783ec304324c08764b47fece5fce (Lingtai-AI/lingtai-kernel main).

Checks run from a clean detached release worktree:

  • python -m compileall -q src ✅
  • python -m pytest -q ran 2,084 tests: 2,080 passed, 3 skipped, and one filesystem-mail watcher test failed once (TestListen.test_listen_detects_multiple_messages) after missing the first of three rapid filesystem events.
  • The single failed watcher test was rerun 5 times and passed 5/5 ✅
  • Focused release-related suites passed: 227 passed across notification sync, system notifications, messaging notification format, filesystem mail, layers email, tool executor, session, and preset manifest tests ✅
  • Final package build and twine check were run during the PyPI publication step.

Known note: the one full-suite failure is treated as a flaky filesystem watcher timing failure because the exact test passed on five immediate reruns and the focused release suites passed.

Compare: v0.11.2...v0.11.3

Assets 2
Loading