v0.21.3

🚀 What's Changed

✨ New Features

• Updated CommandCode & CrofAI model catalogs (41ad2039)
  • New models added:
    • GLM-5.3 Flash & GLM-5.3 (Z.ai) — CommandCode
    • MiniMax M3 (Free) & MiniMax M2.7 (Free) — CommandCode
    • DeepSeek V4 Flash Vision (exp) — CommandCode
    • Qwen 3.8 Flash & Qwen 3.8 27B — CommandCode
    • Gemini 3.7 Flash (Google) — CommandCode
    • Grok 4.6 (xAI) — CommandCode
    • GLM 5.3 & GLM 5.3 Flash (Z.ai) — CrofAI
  • Updated capabilities & pricing for existing models (e.g., DeepSeek V4 Pro/Flash, Qwen 3.7 Max, MiniMax M2.7, Kimi K2.6/K2.5, GLM 5.1/5.2 Fast, MiMo V2.5 Pro, Inkling, Inkling Small, Laguna S 2.1, Grok 4.5, Gemini 3.6 Flash, etc.), including corrected context windows, vision support, and reasoning flags.
  • Removed deprecated models: GLM 4.7, GLM 4.7 Flash (CrofAI), Muse Spark 1.1 (CommandCode).

🐛 Bug Fixes

• Fixed argument preview truncation altering tool outcomes (69d3a049)
  • Dict/list tool arguments are no longer string-truncated in previews; they are now summarized as [an object with typed dict] / [an object with typed list], preventing preview "shrinking" from corrupting or changing the actual tool call.
• Removed the pip-system-certs dependency (94326434)
  • No longer required; slims down the dependency tree and lockfile.

🧹 Improvements & Chores

• Enhanced code-analysis prompting (36e1a9d8)
  • Added a CodeAnalysis_Tool_Instruction prompt guiding agents on tool selection (analyze_repo, find_definition, find_references, grep_text, read_file).
  • Improved find_definition / find_references / grep_text tool descriptions with positive selection guidance ("prefer this over grep_text when you know the exact symbol").
  • Added unit tests covering the new prompt and tool descriptions.