CommandCodeAI / CommandCodeAI/command-code

Feature request: expose hook/plugin system for third-party monitoring tools

未關閉
#604 0 則留言 0 個 reaction 已指派 0 人 在 GitHub 檢視

還沒有人認領這個 Issue。

主要語言
沒有語言資料
星號
4k
分支
350
PR 合併指標
30 天內沒有已合併 PR

描述

The ask

Add a hook or plugin system so third-party apps like AgentPeek can monitor Command Code sessions. AgentPeek currently supports 25 coding agents (Claude Code, Codex, Cursor, OpenCode, etc.) but cannot integrate with Command Code because there is no hook or plugin API to tap into.

What other agents provide

  • Claude Code — hook events surfaced through ~/.claude/settings.json. Hooks fire for tool calls, permission prompts, sub-agents, etc.
  • OpenCode — a plugin system at ~/.config/opencode/plugins/, where AgentPeek drops a JS file to receive session events.
  • Others — at minimum, structured session/transcript/usage data stored on disk in a readable format (JSON, JSONL, SQLite) that tools can tail.

What Command Code would need to expose

At minimum:

  1. Session lifecycle events — session start/end, project path, model, state (thinking, executing, waiting, idle)
  2. Activity events — tool calls, files touched, diffs, commands run
  3. Permission prompts — the command/file/diff being proposed, so tools can surface them in their UI
  4. Usage data — token counts, cost, rate-limit windows (readable from disk or an events stream)

The ideal path is a hook system similar to Claude Code, or a plugin directory like OpenCode. Even just documenting where session data lives on disk would be a good first step.

Use case

Tools like AgentPeek let you monitor all your coding agents from the Mac notch/menu bar — answer permission prompts without finding the terminal, track usage across agents, view session transcripts, etc. Command Code users would benefit from the same visibility.

貢獻指南

這個儲存庫沒有索引到貢獻指南

從這裡開始

  1. 先讀完整個 Issue,再讀專案的貢獻指南。
  2. 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
  3. Fork 儲存庫,在一個分支上完成修改。
  4. 送出 Pull Request,並在描述裡引用這個 Issue 編號。

研究方向

該 issue 未指定任何檔案、測試或進入點。首先定位 Command Code 產生工作階段、活動、權限和使用狀況資料的位置,然後檢視 hook、plugin 或可讀取的事件串流如何公開這些資料。完成的標準是:有一條文件化的整合路徑,讓第三方監控器能夠消費所需的事件。

由索引模型根據 Issue 內容生成。

評估

領域
cli, developer-experience, tooling
Issue 類型
功能
難度
5/5
預估耗時
一週以上
活躍度
冷清
描述清晰度
基本清楚
新手友好度
35/100

把新 issue 寄到你的電子郵件信箱

精選適合新手參與的 GitHub issue 摘要。