v1.0.6 - Support Xiaomi MiMo-v2.6-flash & Real-time Live Settings Polling

Changes in v1.0.6

• Xiaomi MiMo & Modern Reasoning Models Support:
  • Added dedicated rules for \mimo\ models (e.g. \mimo-v2.6-flash, \mimo-v2.6-pro) in \�uiltin-rules.json\ with 1M context window, 131k max tokens, multimodal (text/image/video/audio), tool call, and full reasoning efforts.
  • Enabled
    easoningEfforts\ (\off, \low, \medium, \high, \max) so the reasoning effort selector dropdown displays in DSH Web and Desktop UI.
• Dynamic Real-Time Settings Polling:
  • Implemented lightweight periodic memory signature polling (\getSignature()\ every 2.5s) to automatically detect newly added custom models/routes in DSH UI without requiring application restart.
  • Automatically enriches missing contextWindow, maxTokens, and reasoningEfforts for newly inputted models.

v1.0.5 - Deep-clone Frozen Settings & Cross-platform Log

Changes in v1.0.5

• Fix 'Cannot add property User-Agent, object is not extensible':
  • Deep-cloned settings providers using \structuredClone\ to safely mutate frozen descriptor values returned by Cordis settings service.
  • Added conflict retry handling (\SETTINGS_CONFLICT) on \settings.mutate.
• Cross-platform Diagnostic Path:
  • Implemented \getDshHome()\ to reliably locate ~/.dsh/smart-config.log\ on Windows, Linux, and Android.
• Local DSH Profiles Compatibility:
  • Verified and integrated with local DSH profiles (\web, \desktop, \dsh-tui, \headless).

v1.0.4 - Fix Cordis Property Access on DSH Mobile Startup

Changes in v1.0.4

• Fix Cordis 4 'cannot get property command without inject' error:
  • Removed un-injected \ctx.command\ access causing crash during daemon boot in DeepSeek Harness mobile.
  • Replaced \ctx.logger\ with console logger and diagnostic file logger (\/smart-config.log).
  • Added safe optional service resolver helper to avoid Cordis proxy traps.
• Enhanced Settings Sync:
  • Scheduled immediate and delayed sync passes to ensure model capabilities and reasoning efforts are populated once the settings service is ready.
  • Cleaned up scheduled tasks on plugin dispose.

v1.0.3 - DSH Active Settings Sync, Thinking Effort Adaptation & OpenCode Request Headers

What's Changed in v1.0.3

🌟 DSH Active Settings Sync & Capability Adaptation

• Deep Cordis Integration: Added inject = ['settings'] to deeply integrate into DeepSeek Harness's runtime container.
• Smart Capability Resolver (resolveModelCapabilities):
  • Automatically calculates and injects reasoningEfforts (off, low, high, max), directly illuminating the Thinking Effort selector in DSH Mobile / Web Composer.
  • Automatically adapts model context windows (up to 1,000,000 tokens for GLM-5 and DeepSeek models, 1,048,576 for Kimi).
  • Automatically configures max output tokens (131072 / 384000) and multimodal image input (input: ['text', 'image']).
  • Automatically injects protocol wire compatibility options (supportsReasoningEffort: true, thinkingFormat: "deepseek", requiresReasoningContentOnAssistantMessages: true, maxTokensField: "max_tokens").
• Proactive Lifecycle Sync (syncSettings): Automatically inspects llm-pi-ai provider routes upon plugin load and ready event, and mutates system settings to populate missing models and capabilities.

🛡️ Full-Chain OpenCode Request Header Injection

• Global fetch interceptor automatically inspects and injects compliant OpenCode authentication headers for all *.opencode.ai requests:
  • User-Agent: opencode/1.0.0
  • x-opencode-client: opencode
  • x-opencode-session: <UUID>
• Built-in provider configuration presets automatically declare default OpenCode headers.

📦 Build & Distribution Improvements

• Added "prepare": "npm run build" to support seamless installation directly from Git repository URLs.
• Updated npm package distribution and comprehensive documentation in both English and Simplified Chinese.

Full Changelog: v1.0.2...v1.0.3

v1.0.2 - Updated Badges & Documentation Distribution

🚀 Improvements in v1.0.2

• Updated NPM README to display clean, non-red tracking badges.
• Published to npmjs.com as dsh-plugin-smart-config@1.0.2.
• Synced across npmjs.org and npmmirror.

v1.0.1 - DeepSeek Harness Bundle Layer & Precompiled Dist

🚀 Improvements in v1.0.1

• DSH Bundle Layer Manifest: Added cordis.patch.yml and dsh.bundle.patch in package.json so DeepSeek Harness recognizes it directly as an official profile layer.
• Precompiled JavaScript Distribution: Precompiled TypeScript to dist/ with declaration files (.d.ts) for seamless execution on any Node/Electron runtime without requiring strip-types.
• Published to NPM: dsh-plugin-smart-config@1.0.1 is now live on npm.
• Mounted in Local Profiles: Automatically mounted into local DSH desktop and web profile configurations.

v1.0.0 - Initial Release (ZCode Smart Config Engine for DeepSeek Harness)

🎉 What's New in v1.0.0

The initial release of dsh-plugin-smart-config — a smart model configuration & capability adaptation plugin for DeepSeek Harness (dsh), reverse-engineered and ported from ZCode.

🌟 Key Highlights

• 5-Layer Cascading Recommendation Engine:
  • Global baseline fallback (modelRules: .*)
  • Model family regular expression matching (GLM-5, DeepSeek, Claude, Kimi, MiniMax, etc.)
  • Protocol-specific adaptations (anthropic-messages, openai-chat-completions, openai-responses)
  • Site/Proxy overrides based on normalized baseUrl (OpenCode, DashScope, BigModel, Z.ai)
  • Template & exact provider rule resolution
• Full Official Rule Base (Revision 30):
  • Pre-packaged with 84+ model rules, 72+ API rules, 52+ site rules, and 20+ provider templates.
• Smart Field-Level Diff Overlay:
  • Automatically keeps unedited parameters updated with official recommendations while preserving custom manual overrides.
• Capability Guards & Parameter Mapping:
  • Prevents 400 Bad Request errors by stripping tool schemas from models that do not support function calling.
  • Automatically translates generic reasoning_effort into model-specific JSON structures (DeepSeek enable_thinking, Claude adaptive thinking, OpenAI reasoning_effort).
• Cordis & DSH Integration:
  • Automatically registers the smartConfig service and model/before-call lifecycle hook.
  • Built-in CLI command: smart-config <modelId>.