Newer vLLM parsers put CoT in reasoning; older builds and DeepSeek still use reasoning_content. Canonicalize to the UI field and echo only the original wire field so other ChatOpenAI models stay unchanged. Point qwen3.6 at VllmChatModel.
Co-authored-by: Cursor <cursoragent@cursor.com>
Add 极速 mode that streams via /api/direct-chat without the lead agent stack,
plus VITE_TURBO_MODE_VISIBLE in runtime-config.js (default off). Also disable
global memory, tighten Qwen/vLLM thinking control for flash mode, and force
thinking off on the direct-chat endpoint.
Co-authored-by: Cursor <cursoragent@cursor.com>
Ship report rewrite/version APIs, security budget and audit persistence, AgentScope worker wiring, and the 20-task quality eval harness so the frontend can integrate against a stable contract once enabled in config.
Co-authored-by: Cursor <cursoragent@cursor.com>
Unify Wiki reading, preserve metadata and graph links, add asynchronous file-backed transfers, and reuse compatible vectors. Includes regression coverage and live acceptance notes.
- Strip write_file/str_replace/bash from the model binding in the
first-report run: after_model rewrites only fire after a full model
response, so any bound write tool let weak models draft the whole
essay into tool args (write_file content or bash cat>/python
heredocs) for tens of invisible seconds before interception
- Hallucinated hand-writes are still handled: after_model rewrites
bash cat >/python open('w')/tee and leftover write calls to
deep_research_report; wrap_tool_call rejects whatever reaches the
tool layer; post-success writes are dropped, nudged to a 2-3
sentence summary, and patched with a fallback closing message so
the turn never ends silently
- The rewritten pipeline call carries topic = original request
(clarification answers skipped) plus focus = clarification answers,
so the pipeline writes the clarified subject instead of the raw
prompt text; prompts updated to match (tools unavailable, call the
tool directly, do not draft first)
- Frontend write-step interception and streaming preview scoping now
applies only in writing mode; normal mode shows/previews writes
exactly as before
Keep deep_research_report as a write-only deliverable from conversation materials, block post-success rewrite loops with a summary nudge, fill thin/empty pools from model knowledge, and open the sandbox only on real report.md content.
Co-authored-by: Cursor <cursoragent@cursor.com>
Keep Win7 flowchart SVG labels and bring over Mermaid syntax repair plus HTML chart color downleveling.
Co-authored-by: Cursor <cursoragent@cursor.com>
Rewrite oklch/color-mix and space-separated rgb before iframe preview so pie and bar fills are not dropped on Chrome 109.
Co-authored-by: Cursor <cursoragent@cursor.com>
gozeros