v0.3.0 Release Notes

Released 2026-02-09

v0.2.0 -> v0.3.0

Summary

Add multi-agent backend support (OpenAI, Claude, Gemini, Nvidia) with unified Agent interface architecture, complete filesystem toolchain, and interactive confirmation mechanism.

翻譯

新增多 Agent 後端支援（OpenAI、Claude、Gemini、Nvidia），建立統一 Agent Interface 架構，完善檔案系統工具鏈與互動式確認機制。

Changes

FEAT

• Add OpenAI Agent with API key authentication and gpt-5-nano default model
• Add Claude Agent with Anthropic Messages API integration and typed response deserialization
• Add Nvidia Agent with NIM API support and multiple model options
• Add Gemini Agent with Google Generative AI API and function calling protocol conversion
• Add search_content tool for regex-based file content searching
• Add run_command tool with command whitelist and safe rm via .Trash mechanism
• Add list_files, glob_files, write_file filesystem tools with directory exclusion
• Add interactive confirmation prompt before each tool call with --allow flag to bypass
• Add agent selection menu (GitHub Copilot, OpenAI, Claude, Gemini, Nvidia) at runtime
• Add .env support via godotenv for API key management

翻譯

• 新增 OpenAI Agent，支援 API Key 認證與 gpt-5-nano 預設模型
• 新增 Claude Agent，整合 Anthropic Messages API 與型別化回應反序列化
• 新增 Nvidia Agent，支援 NIM API 與多模型選項
• 新增 Gemini Agent，整合 Google Generative AI API 與 Function Calling 協定轉換
• 新增 search_content 工具，支援正規表示式檔案內容搜尋
• 新增 run_command 工具，含指令白名單與 rm 安全刪除至 .Trash 機制
• 新增 list_files、glob_files、write_file 檔案系統工具與目錄排除機制
• 新增工具呼叫前的互動式確認提示，支援 --allow 旗標跳過確認
• 新增執行時 Agent 選擇選單（GitHub Copilot、OpenAI、Claude、Gemini、Nvidia）
• 新增透過 godotenv 的 .env 支援以管理 API Key

REFACTOR

• Restructure internal/client into internal/agents with unified Agent interface (Send + Execute)
• Extract Copilot implementation into dedicated internal/agents/copilot sub-package
• Extract shared execution loop into agents.Execute() with 128-iteration tool call limit
• Extract OpenAIToolCall as shared type for cross-agent response normalization
• Restructure Skill Scanner with concurrent goroutine scanning and SkillList indexing
• Refactor Claude agent from map[string]any response parsing to typed Output/Content structs
• Embed system prompt via //go:embed sysprompt.md

翻譯

• 重構 internal/client 為 internal/agents，建立統一 Agent 介面（Send + Execute）
• 將 Copilot 實作抽離至獨立 internal/agents/copilot 子套件
• 將共用執行迴圈抽出為 agents.Execute()，限制最多 128 次工具呼叫迭代
• 將 OpenAIToolCall 抽出為共用型別，用於跨 Agent 回應正規化
• 重構 Skill Scanner，採用 Goroutine 併發掃描與 SkillList 索引
• 重構 Claude Agent 從 map[string]any 回應解析改為型別化 Output/Content 結構
• 透過 //go:embed sysprompt.md 嵌入系統提示

CHORE

• Add .env.example with OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, NVIDIA_API_KEY

翻譯

• 新增 .env.example，包含 OPENAI_API_KEY、ANTHROPIC_API_KEY、GEMINI_API_KEY、NVIDIA_API_KEY

REMOVE

• Remove legacy internal/client/copilot.go and internal/client/exec.go
• Remove internal/skill/skill.go (merged into scanner.go)

翻譯

• 移除舊版 internal/client/copilot.go 與 internal/client/exec.go
• 移除 internal/skill/skill.go（合併至 scanner.go）

---

Files Changed

---

Generated by SKILL