LingTai kernel v0.14.0

LingTai kernel v0.14.0

Release window: v0.13.1..v0.14.0 — 39 commits, 75 files changed (+4,919 / −614).

Highlights

• Honest Codex metadata envelope — agents send an accurate metadata envelope to the Codex backend, and the user's own ChatGPT-Account-ID header is forwarded on Codex requests, improving attribution and account routing (#461, #454).
• LLM latency telemetry — provider-wait and per-phase LLM timings are now recorded, giving operators visibility into where turn latency is spent (#366).
• Tool-result attention & agent meta — long tool results are surfaced in agent meta with a top-10 preview list and char counts; metadata is nested under _meta and excluded from summarize sizing so it no longer distorts context budgeting (#451, #452, #453, #465).
• Durable tool-result replay — durable tool results are replayed before heal, hardening post-tool continuation recovery (#297).
• WorkerStillRunning poison recovery — turns poisoned by a stuck worker now recover instead of wedging (#463).
• Molt-pressure config removed — legacy molt-pressure configuration is ignored; molt pressure moved into agent meta (#464, #467).

Other fixes

• Duplicate _check_duplicate_process prefix/substring false-positives fixed (#459).
• Daemon CLI SIGTERM/143 attributed to local cause (#456).
• Notification meta envelope persisted; guidance/meta README rendered as a guidance section (#458, #460).
• Codex thinking preset config repaired (#457).

Validation

• pytest -q: 2635 passed, 4 skipped (273s).
• python -m build + twine check: both artifacts PASSED.

Artifacts

• lingtai-0.14.0-cp311-cp311-macosx_14_0_arm64.whl — 2,137,629 bytes — sha256:37c2ee8978b16bc9273e6240b3832f42b1b008671b3422741f7f0a35e32284ca
• lingtai-0.14.0.tar.gz — 1,295,795 bytes — sha256:279a1ba42e17c6a972390cea67217b0ca125591cf1fa76b4a7890c70e1a70461

Compare: v0.13.1...v0.14.0