Agenvoy v0.32.15 Release Notes
v0.32.14 -> v0.32.15
Summary
Reworks command authorization around deny policies and adds native package management for Linux environments. Makes interactive confirmations channel-aware and improves skill-informed agent selection. Documentation now reflects the revised security and integration behavior.
翻譯
重新設計指令授權流程，改採拒絕政策並為 Linux 環境加入原生套件管理；互動確認現在可辨識來源頻道，Agent 選擇��會參考 Skill 資訊，相關安全與整合文件已同步更新。
⚠️ Breaking Changes
Command authorization now uses deny policies instead of command allowlists
The command allowlist configuration and its management surfaces have been removed. Command execution now evaluates read-only commands, sensitive paths, restricted operations, sandbox boundaries, and explicit deny policies rather than requiring a command to appear in a global allowlist.
Migration:
- Remove integrations that read or modify
white_list.json, call the allowlist API, or invoke the TUI allowlist command. - Move prohibited command behavior into deny-policy checks and keep non-mutating commands in the read-only command configuration.
- Review custom execution integrations to ensure sensitive paths and restricted operations still require confirmation.
- Use the package-management tool for privileged Linux system package operations instead of relying on allowlisted package-manager commands.
翻譯
指令授權已移除全域允許清單設定與管理介面，改由唯讀指令、敏感路徑、受限操作、sandbox 邊界及明確拒絕政策共同判定。
遷移方式：
- 移除讀寫
white_list.json、呼叫允許清單 API 或使用 TUI 允許清單指令的整合。 - 將禁止執行的指令規則移至拒絕政策，非變動性指令則維護於唯讀指令設定。
- 檢查自訂執行整合，確保敏感路徑與受限操作仍會要求確認。
- Linux 系統套件的提權操作改用套件管理工具，不再依賴允許清單中的套件管理指令。
Pending confirmations are routed by execution origin
Pending confirmation requests now carry an execution origin and are claimed by the matching CLI, web, Telegram, or Discord listener. The TUI no longer consumes pending requests from every channel, and confirmation waits now use a bounded timeout.
Migration:
- Set the execution origin when constructing execution metadata for a non-CLI channel.
- Register pending listeners with the channel prefix they own instead of using an unfiltered listener.
- Pass the new skill hint argument when calling agent-selection functions; use an empty string when no skill context exists.
- Handle confirmation timeout as a skipped tool execution and allow the original channel to present the result.
翻譯
待確認請求現在會攜帶執行來源，並由對應的 CLI、Web、Telegram 或 Discord listener 接收；TUI 不再取得所有頻道的待確認請求，等���確認也改為有限逾時。
遷移方式：
- 非 CLI 頻道建立執行 metadata 時，必須設定執行來源。
- 待確認 listener 應使用其負責的頻道前綴註冊，不再使用未篩選 listener。
- 呼叫 Agent 選擇函式時傳入新的 Skill 提示參數；沒有 Skill 上下文時使用空字串。
- 將確認逾時視為略過工具執行，並由原始頻道呈現結果。
Changes
BREAKING
- Replace command allowlists with deny policies, expand read-only command detection, and add privileged Linux package management (@pardnchiu) [eb71633]
- Route pending confirmations by execution origin and pass skill descriptions into agent selection (@pardnchiu) [8da8544]
翻譯
- 以拒絕政策取代指令允許清單、擴充唯讀指令判定，並新增需提權的 Linux 套件管理功能 (@pardnchiu) [eb71633]
- 依執行來源路由待確認請求，並將 Skill 說明傳入 Agent 選擇流程 (@pardnchiu) [8da8544]
FEAT
- Unify sensitive-path enforcement and stream command progress across interactive clients (@pardnchiu) [2fc9962]
翻譯
- 統一敏感路徑限制，並在互動式客戶端串流顯示指令進度 (@pardnchiu) [2fc9962]
DOC
- Update documentation for current CLI, API, security, and integration behavior (@pardnchiu) [27c54ef]
- Align documentation with skill-aware routing, channel confirmations, and package management (@pardnchiu) [1621285]
翻譯
- 更新目前 CLI、API、安全與整合行為的文件 (@pardnchiu) [27c54ef]
- 使文件與 Skill 感知路由、頻道確認及套件管理行為一致 (@pardnchiu) [1621285]
CHORE
- Remove obsolete chatbot client methods and MCP health APIs (@pardnchiu) [4a2fa3a]
翻譯
- 移除已停用的聊天機器人 client 方法與 MCP 健康檢查 API (@pardnchiu) [4a2fa3a]
Scope
configs/jsons/— BREAKING, FEATinternal/agents/exec/— BREAKING, FEATinternal/filesystem/— BREAKING, FEATinternal/runtime/— BREAKING, FEAT, CHORE (multiple files)internal/tools/— BREAKING, FEAT (multiple files)page/src/js/— BREAKING, FEATREADME.md— DOCdoc/— DOCextensions/skills/— DOC, CHORE
Generated by SKILL