v0.5.0: Dynamic Model Discovery, OpenCode 2 Support, and Plan Mode Context

Highlights in v0.5.0

• ⚡ Real-time Model Discovery: Automatically fetches active models and context limits from Command Code's public endpoint (https://api.commandcode.ai/provider/v1/models) on startup with zero-credential requirement and instant offline fallback.
• 🚀 OpenCode 2 (V2) Support: Native ./v2 plugin entrypoint (plugins: ["commandcode-go-opencode-provider/v2"]) with automatic catalog transformation — resolves SchemaError from upstream.
• 🧭 Plan Mode Project Context: Injects project file tree (respecting .gitignore) and Git repository metadata (current branch, status, recent commits) with 30s TTL in-memory caching.
• 🧠 Thinking & Reasoning Effort: Full support for custom thinking levels (low, medium, high, max) and token budgets.
• 📦 67 Verified Models: Offline catalog expanded to 67 models with case-sensitive IDs, accurate capabilities, and context limits up to 1M tokens.
• 🛡️ Production-Grade Resilience: Exponential backoff with ±25% jitter, error classification (18 permanent vs 24 retryable), mid-stream reconnect protection, and clean SSE stream terminations.

Quality Assurance

• 106 unit tests passing (100% pass rate)
• Strict TypeScript typechecking (tsc --noEmit) passing with 0 errors

v0.4.1: Production-Grade Retry/Backoff and 50 Models Schema

Highlights in v0.4.1

• Retry Engine with Jitter: Fixed backoff schedule [1s, 2.5s, 5s] with ±25% jitter.
• Error Classification: 18 non-retryable patterns (auth/quota/validation) and 24 retryable patterns.
• Mid-stream Reconnect Protection: emittedContent tracking to prevent duplicate outputs.
• SSE Error Propagation: Terminate streams cleanly via controller.error() instead of silent swallowing.
• Curated 50 Models: Added full OpenCode model schema (capabilities, vision, PDF, reasoning variants) for 50 Command Code models.