Token导航 LogoToken导航TokenDH.com
研究检索敏感数据github未标认证来源可访问许可证需确认审计异常

virtuivirtui 搜索

Agent Skill

virtui 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

212

周安装

9

GitHub Stars

82

下载量

74
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:virtui(virtui 搜索)
来源仓库:https://github.com/honeybadge-labs/virtui
仓库路径:skills/virtui
安装命令:
npx skills add https://github.com/honeybadge-labs/virtui --skill virtui
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/honeybadge-labs/virtui --skill virtui

简介

用于查找、检索和筛选相关信息。virtui 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

  • 适合根据关键词或任务场景快速定位候选结果。
  • 可结合来源仓库和原始 README 核验具体用法。
  • 安装前建议确认权限范围和维护状态。
  • 注意是否会触发联网或文件读写。适用宿主包括 Codex、Claude、Cursor、Gemini CLI,接入前应确认版本、权限和运行环境要求。

SKILL.md

virtui — TUI Automation for AI Agents

Prerequisites

Before using virtui, ensure it is installed and the daemon is running.

Check / Install

# Check if virtui is available
virtui version

If not installed, install via Homebrew:

brew install honeybadge-labs/tap/virtui

Start the Daemon

The daemon must be running before any session commands. Always check first:

virtui --json daemon status
# → {"running":true,"socket":"..."} or {"running":false,"socket":"..."}

If not running:

virtui --json daemon start
# → {"pid":1234,"socket":"/Users/you/.virtui/daemon.sock"}

The daemon manages sessions over a Unix socket at ~/.virtui/daemon.sock.

Core Workflow

The typical workflow is: start daemon → run session → interact → screenshot → kill session.

1. Launch a Session

virtui --json run <command...>

This returns a session_id you'll use for all subsequent commands. Common options:

FlagDefaultPurpose
--cols80Terminal width
--rows24Terminal height
--dircwdWorking directory
-e KEY=VALEnvironment variables (repeatable)
--recordoffRecord session as asciicast v2
--record-pathautoCustom recording path

Example:

virtui --json run --cols 120 --rows 40 bash
# → {"session_id":"a1b2c3d4","pid":1234,"recording_path":""}

2. Execute Commands

exec is the primary interaction command. It types input, presses Enter, and optionally waits for a screen condition before returning.

Claude Code note: The Bash tool strips trailing newlines from command arguments, so exec may not actually send Enter when called from Claude Code. Prefer using type + press Enter as separate steps (or use a pipeline) to guarantee Enter is sent. See the Pipeline section for a concrete example.
Wait-condition caveat: Wait conditions check the screen immediately after input is sent. If the target text already appears on screen (e.g., in the typed command itself), the wait can resolve in 0 ms — before the command's actual output appears. For reliable results, use a pipeline with separate typepress Enterwait steps, or follow exec with a standalone wait command.
virtui --json exec <session_id> "<input>" [wait flags]

Wait strategies (use exactly one, or none to return immediately with current screen state):

FlagWhen to use
--wait "text"Wait for specific text to appear on screen
--wait-stableWait for screen to stop changing (500 ms of no updates — does not guarantee the process finished)
--wait-gone "text"Wait for text to disappear (e.g., a loading spinner)
--wait-regex "pattern"Wait for a regex match on screen
--timeout <ms>Override default 30s timeout (use with any wait flag)

Examples:

# Run a command and wait for its output
virtui --json exec $SID "echo hello" --wait "hello"

# Run a build and wait for screen to settle (500ms of no changes)
virtui --json exec $SID "make build" --wait-stable --timeout 60000

# Wait for a loading indicator to disappear
virtui --json exec $SID "npm install" --wait-gone "resolving" --timeout 120000

3. Take Screenshots

Read the current screen contents at any time:

virtui --json screenshot <session_id>
virtui screenshot <session_id> --no-color

Returns screen_text, screen_hash, screen_ansi, cursor position, and terminal dimensions. Use screen_hash for cheap change detection without comparing full screen text.

  • Use screen_text for plain text assertions and screen-diff logic.
  • Use screen_ansi when color, background color, bold, underline, or reverse video matter.
  • Use virtui screenshot --no-color <session_id> when you want plain text output in non-JSON mode without ANSI escape sequences.

4. Send Keystrokes

For interactive TUI applications (menus, editors, prompts), use press and type:

# Send special keys
virtui press <session_id> Enter
virtui press <session_id> Ctrl+C
virtui press <session_id> ArrowDown --repeat 5
virtui press <session_id> Escape q      # multiple keys in sequence

# Type text without pressing Enter (for search fields, partial input)
virtui type <session_id> "search query"

See references/keys-and-errors.md for the full list of available key names and error codes.

5. Wait for Conditions

Wait independently of exec (useful after press/type or for polling):

virtui --json wait <session_id> --text "Ready"
virtui --json wait <session_id> --stable
virtui --json wait <session_id> --gone "Loading..."
virtui --json wait <session_id> --regex "v\d+\.\d+"

6. Clean Up

Always kill sessions when done:

virtui kill <session_id>

List active sessions:

virtui --json sessions show
# → {"sessions":[{"session_id":"a1b2c3d4","pid":1234,"command":["bash"],"cols":80,"rows":24,"running":true,"exit_code":-1,"created_at":"1711900000","recording_path":""}]}

7. Resize a Session

Resize the terminal dimensions of a running session. Both --cols and --rows are required (unlike run, there are no defaults):

virtui resize <session_id> --cols 120 --rows 40

Pipeline (Batch Operations)

For complex multi-step interactions, use pipeline to send a batch of steps in one call.

Recommended for Claude Code: Because exec relies on Enter being sent and Claude Code's Bash tool can swallow trailing newlines, the most reliable pattern from Claude Code is to use a pipeline with explicit type + press Enter steps instead of exec.

Example: Running a command reliably from Claude Code

echo '{"steps":[
  {"type":{"text":"echo hello world"}},
  {"press":{"keys":["Enter"]}},
  {"wait":{"condition":{"text":"hello world"},"timeout_ms":5000}},
  {"screenshot":{}}
],"stop_on_error":true}' | virtui --json pipeline $SID

This is equivalent to virtui exec $SID "echo hello world" --wait "hello world" but guarantees Enter is actually sent regardless of how the shell tool handles newlines.

Pipeline from file

Instead of piping JSON via stdin, you can pass a file containing the steps:

virtui --json pipeline <session_id> --file steps.json

General pipeline example

echo '{"steps":[
  {"exec":{"input":"ls","wait":{"text":"README"}}},
  {"sleep":{"duration_ms":500}},
  {"screenshot":{}},
  {"press":{"keys":["Ctrl+C"]}}
],"stop_on_error":true}' | virtui --json pipeline <session_id>

Step types: exec, press, type, wait, screenshot, sleep.

Pipeline output format

The pipeline returns a JSON object with a results array. Each entry contains the step outcome and a screenshot of the screen state at that point:

{
  "results": [
    {
      "step_index": 0,
      "success": true,
      "error_message": "",
      "screenshot": {
        "screen_text": "...",
        "screen_hash": "...",
        "screen_ansi": "...",
        "cursor_row": 4,
        "cursor_col": 10,
        "cols": 80,
        "rows": 24
      }
    }
  ]
}

Recording Sessions

Record sessions as asciicast v2 (playable with asciinema play):

virtui --json run --record bash
# or with a custom path:
virtui --json run --record --record-path ./demo.cast bash

Recording stops when the session is killed or the process exits.

Note: --record-path requires --record to be set. Using --record-path alone will not enable recording.

Important Patterns

  • Always pass --json (or -j) for machine-readable output with session_id, screen_text, screen_hash, screen_ansi, etc.
  • JSON output uses proto3 JSON encoding: int64 fields (elapsed_ms, created_at) are serialized as strings.
  • Use screen_hash (SHA-256) for cheap change detection without comparing full screen text.
  • Prefer screen_text for text matching and screen_ansi only when TUI state depends on styling or color.
  • The default wait timeout is 30s. For long-running ops, increase it: --timeout 120000.
  • If a wait times out, take a screenshot to see current state and decide how to proceed.
  • Errors include code, category, message, retryable, suggestion — check retryable before retrying. See references/keys-and-errors.md for error codes and key names.

JSON Output Fields Reference

CommandKey Fields
runsession_id, pid, recording_path
execscreen_text, screen_hash, cursor_row, cursor_col, elapsed_ms
screenshotscreen_text, screen_hash, screen_ansi, cursor_row, cursor_col, cols, rows
pressscreen_text, screen_hash
typescreen_text, screen_hash
waitscreen_text, screen_hash, elapsed_ms
killok
resizeok
pipelineresults[]: step_index, success, error_message, screenshot (see Pipeline output format)
sessionssessions[]: session_id, pid, command, cols, rows, running, exit_code, created_at, recording_path
daemon startpid, socket (background) or socket (foreground)
daemon stopok, optionally message
daemon statusrunning, socket

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

33.38%
按下载量换算25

Claude

30.91%
按下载量换算23

Cursor

18.58%
按下载量换算14

Gemini CLI

9.92%
按下载量换算7

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

未通过

权限和风险

敏感数据

该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在 warning/failed 结果,不能写成本站确认安全。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills