将 SessionStart Hook 的 stdout 展示给用户 || Display the stdout of SessionStart Hook to the user

What feature would you like to see?

功能请求：将 SessionStart Hook 的 stdout 展示给用户

问题

目前 SessionStart hook 执行后会捕获 stdout/stderr，但不会展示给用户。这限制了 hook 在以下场景中的实用性：

• 进入工作空间时显示欢迎语 / 项目仪表盘
• 在用户开始输入前展示项目健康度/状态
• 运行快速诊断并立即展示结果

使用场景

我的项目使用 AGENTS.md 文件，其中有一条规则：

"AI 进入时，必须先读取 WELCOME.md，向 owner 展示问候语和菜单，再等待输入。"

我尝试配置了一个 SessionStart hook：

[[hooks]]
event = "SessionStart"
command = "cat WELCOME.md"
timeout = 5

Hook 执行成功了，但输出被 hook 引擎吞掉了，用户永远看不到。我只能退而求其次用 shell 别名来包装，但这终究是个 workaround。

建议方案

提供一种机制，将 hook 的 stdout 输出到用户界面。有几种可能的实现方式：

方案 A：直接展示所有 SessionStart hook 的 stdout（最简单）

当 SessionStart hook 执行时，在进入交互式提示符之前，直接把它的 stdout 打印到终端。

方案 B：在 hook 配置中增加 display_output 开关（推荐）

[[hooks]]
event = "SessionStart"
command = "cat WELCOME.md"
timeout = 5
display_output = true   # 新增：将 stdout 透传给用户界面

向后兼容，默认关闭，显式开启。

方案 C：支持结构化输出并声明展示意图

允许 hook 返回一个 JSON 响应，指明哪些内容需要展示：

{
  "action": "allow",
  "display": {
    "text": "...hook 的 stdout 内容...",
    "format": "markdown"
  }
}

为什么不用 Shell 别名？

Shell 包装确实能用，但：

1. 不便携 — 团队成员需要各自配置别名
2. 丢失上下文 — Hook 在 kimi 的工作目录上下文中执行（通过 stdin 接收 cwd、session_id 等 JSON），而 shell 别名做不到
3. 无法利用 matcher — Shell 别名每次必跑；kimi hook 支持正则 matcher，可按条件触发
4. 像是 workaround — 既然 hooks 系统已经如此强大，理应原生支持这种场景

环境

• kimi-cli 版本: 1.30.0
• 操作系统: Ubuntu 22.04
• 配置文件: ~/.kimi/config.toml

相关代码

Hook 的执行逻辑在 kimi_cli/hooks/runner.py 中，stdout 当前仅用于：

• 退出码 2 → block 操作
• 退出码 0 + JSON → 结构化权限决策

stdout 字符串在 HookResult.stdout 中已有，但从未被转发到用户界面层。

---

你们是否接受针对方案 B（opt-in display_output 开关）的 PR？ 如果方案被认可，我很乐意实现并提交。

Additional information

No response

---

What feature would you like to see?

Feature request: Display the stdout of SessionStart Hook to the user

Question

Currently, SessionStart hook will capture stdout/stderr after execution, but will not be displayed to the user. This limits the hook's usefulness in the following scenarios:

• Show welcome/project dashboard when entering workspace
• Show project health/status before user starts typing
• Run quick diagnostics and display results instantly

Usage scenarios

My project uses the AGENTS.md file, which has a rule:

"When the AI enters, it must first read WELCOME.md, show the greeting and menu to the owner, and then wait for input."

I tried configuring a SessionStart hook:

[[hooks]]
event = "SessionStart"
command = "cat WELCOME.md"
timeout=5

Hook execution is successful, but the output is swallowed by the hook engine and the user will never see it. I had to fall back on wrapping it with a shell alias, but this is a workaround after all.

Suggestions

Provides a mechanism to output the hook's stdout to the user interface. There are several possible implementations:

Solution A: Directly display the stdout of all SessionStart hooks (the simplest)

When the SessionStart hook executes, print its stdout directly to the terminal before entering the interactive prompt.

Solution B: Add the display_output switch in the hook configuration (recommended)

[[hooks]]
event = "SessionStart"
command = "cat WELCOME.md"
timeout=5
display_output = true # New: Transparently pass stdout to the user interface

Backward compatibility, off by default, enabled explicitly.

Solution C: Support structured output and declare display intent

Allow the hook to return a JSON response indicating what content needs to be displayed:

{
  "action": "allow",
  "display": {
    "text": "...hook's stdout content...",
    "format": "markdown"
  }
}

Why not use Shell aliases?

Shell wrapping does work, but:

1. Not portable — Team members need to configure their own aliases
2. Lost context — Hook is executed in the context of kimi's working directory (receiving cwd, session_id, etc. JSON via stdin), while shell aliases cannot do this
3. Unable to use matcher - Shell alias must be run every time; kimi hook supports regular matcher and can be triggered according to conditions
4. Like a workaround — Since the hooks system is so powerful, it should natively support this scenario

Environment

• kimi-cli version: 1.30.0
• Operating system: Ubuntu 22.04
• Configuration file: ~/.kimi/config.toml

Related code

The execution logic of Hook is in kimi_cli/hooks/runner.py. stdout is currently only used for:

• Exit code 2 → block operation
• Exit code 0 + JSON → Structured permission decisions

The stdout string is present in HookResult.stdout but is never forwarded to the user interface layer.

---

**Do you accept PRs for option B (opt-in display_output switch)? ** If the proposal is approved, I will be happy to implement it and submit it.

Additional information

No response

Start with kimi_cli/hooks/runner.py and trace HookResult.stdout toward the user interface layer. Clarify whether the accepted design is opt-in display_output or another approach, then define how SessionStart output should appear while preserving existing exit-code and JSON decision handling. Done means the chosen behavior is covered and existing hook semantics remain intact.