should_replay_reasoning_content_for_provider() returned false whenever
provider_accepts_reasoning_content(provider) was false (true for
ApiProvider::Openai) without checking the model. This single gate feeds
both build_for_provider (include_reasoning) and
sanitize_thinking_mode_messages, so a DeepSeek reasoning model on the
generic openai provider (DeepSeek-compatible endpoint) had all
reasoning_content stripped -> the DeepSeek thinking-mode API 400s
('reasoning_content in the thinking mode must be passed back'). This is
the over-aggressive half of ac01b225 (fix#1542).
Gate the early return on the model too:
!provider_accepts_reasoning_content(provider) && !requires_reasoning_content(model).
Known DeepSeek reasoning models replay regardless of provider; genuine
non-DeepSeek models on openai still strip (effort=off still wins). #1542
not regressed (provider_accepts_reasoning_content untouched).
Two pre-existing client.rs tests asserted the buggy case (deepseek-v4-pro
on Openai -> dropped); retargeted to gpt-4o to preserve their #1542
intent without encoding the bug. New positive/negative coverage in
chat.rs.
Refs #1739, #1694, #1542, #1736.
Address gemini-code-assist review on PR #1740 (MEDIUM, clarity): the
non-DeepSeek else-branch match listed ApiProvider::Deepseek /
DeepseekCN arms that are logically unreachable (handled by the if
branch above). Collapse to a single
'Deepseek | DeepseekCN => unreachable!(...)' arm so the intent is
explicit for future maintainers. No behavior change.
Refs #1714.
The CLI --model handoff only exports DEEPSEEK_MODEL, which apply_env_overrides
funneled into the DeepSeek-only root default_text_model slot. default_model()
then treated it as a normalizable weak default and fell back to a
DeepSeek/provider default for an unrecognized custom model on a non-DeepSeek
provider, so e.g. --provider openai --model MiniMax-M2.7 silently sent a
DeepSeek model to the endpoint.
Route DEEPSEEK_MODEL into the active provider's model slot for non-DeepSeek
providers (mirroring the existing OPENAI_MODEL branch), and return an explicit
non-DeepSeek-alias model verbatim from default_model(). DeepSeek/DeepSeekCN
behavior is unchanged. Adds a regression test next to
openai_provider_accepts_custom_model_and_base_url.
Refs #1714, #1739, #1736.
Only DEEPSEEK_LOG_LEVEL should gate verbose CLI output. RUST_LOG controls
the tracing subscriber independently (file logging). On Windows stderr is
not redirected to the log file, so coupling the two causes tracing log
messages to leak into the TUI alt-screen, corrupting the display.
Closes#1774
On Linux, `arboard::Clipboard::new()` opens a blocking connect() to the
X11 Unix socket. When no X server is running (headless, WSL2 without
WSLg), the call hangs indefinitely. Because raw mode and the alternate
screen are already active at that point, Ctrl+C no longer generates
SIGINT and the event loop hasn't started yet — leaving the user with a
blank screen and no way to exit.
Move clipboard initialization from `ClipboardHandler::new()` (called
synchronously during App construction) to a lazy `ensure_clipboard()`
that runs on first read/write with a 500 ms timeout. If the X11
connection doesn't respond in time, the handler stays in fallback mode
and `write_text` falls through to the existing OSC 52 / pbcopy /
PowerShell paths.
The v0.8.38 upgrade dramatically changed two user-visible behaviors that
were not intended as regressions:
- The /model picker was reworked (#1201/#1632) to make a blocking network
fetch on open and replace the curated tier list with the raw provider
catalog. Revert model_picker.rs and the OpenModelPicker handler to the
v0.8.37 instant curated picker. The /models command still lists the live
catalog.
- #1617 rekeyed the approval cache to an exact full-argument fingerprint,
which also dropped the v0.8.37 arity-aware command-family grouping for
"approve for session". Reintroduce build_approval_grouping_key (the lossy
v0.8.37 logic) for approvals while keeping the exact key for denials, so
denying one call no longer over-blocks later differing calls.
https://claude.ai/code/session_01NDuRxM56o17SE7SDLcTFYT
When an OpenAI-compatible backend (vLLM, Ollama, LM Studio, Together AI,
self-hosted vLLM/SGLang, etc.) streams an assistant message containing
multiple tool_calls in a single round, only the **last** tool's
`Event::ToolCallStarted` was firing. The preceding N-1 tool calls
executed and produced tool_result events, but never announced their
start to consumers (TUI / runtime API / embedder bridges), leaving them
with N orphan tool_result blocks and no matching tool_use blocks in the
assistant history.
## Reproduction
```text
backend dispatches: 7 × write_file + 1 × exec_shell
log shows: 7 × ApprovalRequired events ✓
listeners receive: 1 × chat:tool_start, 7 × chat:tool_end
session history: 1 tool_use + 7 tool_result (6 orphans)
```
Tested against vLLM 0.7 + Qwen3.6-35B-A3B with a "scaffold 7-file Tauri
template" prompt. Any model+backend combo that emits batch tool_calls
trips this — typical when a single LLM round asks for multiple parallel
file writes or edits.
## Root cause
`run_turn` tracked the currently-streaming tool block with a single
`current_tool_index: Option<usize>`. The Anthropic-style adapter
(non-streaming response → events at `chat.rs::L1807`) emits
Start/Stop pairs in lockstep so the slot never overlaps. But the
OpenAI streaming parser (`chat.rs::L1954-2064`) emits every
`ContentBlockStart::ToolUse` as soon as a tool_call delta lands, then
batches every `ContentBlockStop` at `finish_reason`:
```text
Start { index: 0 } // tool #1
Delta { index: 0, .. }
Start { index: 1 } // tool #2 — overwrites current_tool_index
Delta { index: 1, .. }
…
Start { index: 6 } // current_tool_index = Some(6)
Delta { index: 6, .. }
Stop { index: 0 } // take() returns Some(6) ← wrong tool!
Stop { index: 1 } // take() returns None
Stop { index: 2 } // take() returns None
…
```
The first `Stop` consumes the last index and emits `ToolCallStarted`
for the wrong `tool_uses` entry; every subsequent `Stop` finds the
slot already `None` and skips the entire `if let Some(index) = …`
branch, dropping the announcement.
## Fix
Replace the single slot with `HashMap<u32 block_index, usize
tool_uses_idx>`:
- `ContentBlockStart::ToolUse` and `::ServerToolUse` insert the
`(event.index → tool_uses.len())` mapping.
- `InputJsonDelta` looks up by the `ContentBlockDelta` outer index.
- `ContentBlockStop` removes by the stop's index, so each Stop routes
to its own `tool_uses` entry regardless of arrival order.
Routing no longer depends on `current_block_kind` (which has the same
single-slot overwrite problem); `current_tool_indices.remove(&index)`
returning `Some(_)` already proves the Stop belongs to a tool block.
## Tests
Added `batch_tool_calls_preserve_all_tool_use_indices` in
`core/engine/turn_loop.rs::tests` — feeds 7 Starts and 7 Stops through
the same `HashMap` API used by `run_turn`, asserts every index round-trips.
Manual end-to-end verification: vLLM + Qwen3.6-35B + 7-file Tauri
template prompt → frontend `messages` history now contains all 7
`write_file` tool_use blocks paired with their tool_result blocks.
Co-authored-by: hexin <he.xin@h3c.com>
Add a release follow-up job that updates the Homebrew tap from the checksum manifest when a tap token is configured.
The job now skips before checkout/download/update when neither HOMEBREW_TAP_PAT nor RELEASE_TAG_PAT is configured, so missing tap credentials do not fail an otherwise successful release.
Closes#1602.
Co-authored-by: Zhiping <2716057626@qq.com>
Co-authored-by: Oliver-ZPLiu <47081637+Oliver-ZPLiu@users.noreply.github.com>
Map legacy DeepSeek CN provider names back to the canonical Deepseek provider in both manual parsing and TOML deserialization.
Co-authored-by: qiyan233 <qiyan233@users.noreply.github.com>
Hard-wrap overlong CJK/no-whitespace runs in diff and pager text wrappers so they do not overflow the right edge.
Fixes#1571.
Co-authored-by: Aitensa <1900013029@pku.edu.cn>
Deduplicate official DeepSeek model completions and normalize known prefixed aliases to the bare model IDs expected by official DeepSeek providers, while preserving provider-specific IDs for compatible backends.\n\nFixes #1594.\n\nCo-authored-by: reidliu41 <reid201711@gmail.com>
Make the translation client optional so missing or invalid API configuration does not crash startup before onboarding can render.\n\nCo-authored-by: Crvena <kuragectl@gmail.com>
Capture and replay Mcp-Session-Id for Streamable HTTP transports, and apply configured custom headers to the GET preflight.\n\nCloses #1629.\n\nCo-authored-by: Zhiping <2716057626@qq.com>
Write the Kitty keyboard protocol probe (ESC[>0u) on Windows instead of enabling disambiguation flags that crossterm does not decode there. Fixes#1599.
Make the sidebar expiry test avoid subtracting from a fresh Instant on Windows runners, falling back to a short sleep only when an older Instant cannot be represented.
Hard-break oversized streaming tokens so CJK runs, URLs, and other no-whitespace content stay within the target width. Includes regression coverage for long CJK text and first-token overflow.
Add the Feishu/Lark long-connection bridge, Tencent Lighthouse runbooks, CNB mirror guidance, CNB tag release pipeline, and China-friendly update fallback documentation for the v0.8.37 line.
Summary:
- restore auto sidebar focus when Ctrl+Alt+0 is pressed from hidden state
- preserve existing hide behavior from visible sidebar states
- add regression coverage
Validation: CI green before merge.
Summary:
- include generic tool input in approval-cache fingerprints
- keep exact repeat denials stable
- prevent one denied generic tool call from blocking future distinct calls
Validation: CI green before merge.
Summary:
- concise live shell/tool labels
- collapsed pending CI polling rows
- hardened stale task-panel timing test
Validation: CI green before merge.
Zhongyue Lin