feat(aihubmix): align with Anthropic official plugin v0.3.24

[DDU1222]

Summary

• Adaptive Thinking: support thinking.type=adaptive for opus-4-7/4-8, sonnet-5, fable-5 (always-on); auto-remove temperature/top_p for adaptive models
• output_config: effort (low~max) + task_budget (≥20000 with beta header task-budgets-2026-03-13)
• Prompt Caching TTL: 5m (default, 1.25x write) / 1h (2.0x write) with per-tier billing in both streaming and non-streaming paths
• 7 bug fixes: assistant message content loss (elif→collect both), multi-tool stream param cross-contamination (str→dict), redacted_thinking close tag, JSON prefill adaptive guard, count_tokens adaptive branch, streaming TTL cache-split lookup, token fallback zero-value bug
• 13 new models: claude-opus-4-8(-think), claude-sonnet-5, claude-fable-5, gpt-5.6-sol/luna/terra, grok-4.5, glm-5.2, hy3, doubao-seed-2-1-pro/turbo, grok-build-0.1
• 7 YAML updates: context_size corrections (opus-4-6/4-7: 200K→1M, haiku-4-5: 204800→200000), document feature, prompt_caching params

Test plan

• Python syntax check (ast.parse) — passed
• All 189 YAML files validated against AIModelEntity schema — 0 failures
• API direct test: adaptive thinking (opus-4-7), always-on (fable-5), disabled (sonnet-5)
• API direct test: output_config effort + task_budget with beta header
• API direct test: cache TTL 1h with per-tier cache_creation breakdown
• API direct test: multi-tool call (2 tools, no param cross-contamination)
• API direct test: gpt-5.6-sol new model
• API direct test: opus-4-6 regression (non-adaptive model unchanged)
• Local Dify community edition: plugin registration, model list, basic conversation
• Line-by-line diff with Anthropic official plugin v0.3.24 — confirmed aligned

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request introduces support for several new models (including Claude Fable 5, Claude Opus 4.8, Claude Sonnet 5, GPT 5.6 variants, Grok 4.5, Doubao, GLM, and Hunyuan 3) and updates existing configurations. It also updates the Anthropic integration to support adaptive thinking, task budgets, effort parameters, and custom prompt caching TTLs. Critical feedback includes resolving thread-safety issues with request-specific state stored on a shared singleton, correcting the prompt token adjustment formula to prevent over-billing, properly handling list-based system prompt serialization, and filtering out empty text blocks in assistant messages to avoid API failures.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.