Token导航 LogoToken导航

Agent Skill · 研究检索

godot-interactive(戈多互动)

godot-interactive 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。

godot-interactive 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为操作浏览器;许可证信息还不明确,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 69,749。

总安装 216周安装 9GitHub Stars 10下载 72核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

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:godot-interactive(戈多互动)
来源仓库:https://github.com/alexmeckes/godot-claude-skills
仓库路径:skills/godot-interactive
安装命令:
npx skills add https://github.com/alexmeckes/godot-claude-skills --skill godot-interactive
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/alexmeckes/godot-claude-skills --skill godot-interactive

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

godot-interactive 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中根据关键词、任务场景或来源线索快速定位候选结果。适用于戈多互动类游戏或交互逻辑相关的技术调研与资料搜集。通过 npx skills add 命令从指定仓库安装,需确保宿主平台兼容。

安装前应检查仓库维护状态与安全策略,避免引入不可靠组件。该技能可能涉及外部查询或文件操作,建议先在小范围测试再推广使用。

来源仓库内路径:skills/godot-interactive

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/alexmeckes/godot-claude-skills --skill godot-i…;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要操作浏览器,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

godot-interactive 怎么安装?

在终端执行 npx skills add https://github.com/alexmeckes/godot-claude-skills --skill godot-interactive。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

godot-interactive 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

godot-interactive 需要什么权限?

权限级别是操作浏览器。该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

godot-interactive 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

godot-interactive 用的人多吗?

公开数据是累计安装 216 次,近一周安装 9 次,下载 72 次,GitHub 10 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

godot-interactive 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 第三方扫描已通过,常见的高风险行为没有命中。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
godot-interactive

戈多互动 · 本页

21610操作浏览器未明确通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Godot Interactive

Use a persistent godot-mcp session to work on Godot 4.x projects without rediscovering the editor state every turn. This skill is the Godot analogue of playwright-interactive: it keeps one live editor connection alive when possible, combines it with file-based scene/script/shader tools, and requires explicit validation evidence before signoff.

Prefer the smallest evidence-producing loop. For live-editor tasks, stay on the live godot_editor_* and godot_runtime_* surface unless a file-only detail is truly required.

Preconditions

  • godot-mcp must be configured for the current project.
  • For live editing, Godot 4.x must be running with the target project open and the AI Bridge plugin enabled.
  • The bridge target should come from the active godot_mcp MCP configuration. 127.0.0.1:6550 is a common default, but local setups may override it.
  • For automated gameplay input or screenshots, prefer the first-class godot_runtime_* tools when they are present. If they are absent, fall back to a project-provided runtime harness or manual verification.
  • If tool choice is unclear, call godot_help first.
  • If the bridge is unavailable, continue with file-based tools and manual Godot review instead of inventing live-editor results.
  • Treat a dropped bridge connection as a session problem. Reconnect before doing more live-editor work.
  • If the user explicitly targets the running editor or currently open scene, treat the connected editor project as the source of truth. Do not burn time reconciling it with the shell workspace unless the task also depends on workspace files or the user suggests the wrong editor is open.

Workflow Decision Tree

  • Use the live-editor workflow for node hierarchy changes, layout tuning, quick property edits, opening/running scenes, and immediate runtime inspection.
  • Use the file-based workflow for larger structural edits, generated scenes/scripts/shaders/resources, or any work done before Godot is open.
  • Use the runtime-automation workflow whenever dedicated godot_runtime_* tools exist. If they do not, only use it when the project already exposes a real runtime harness.
  • Use the hybrid workflow for most real tasks: edit files, refresh the editor, inspect the scene tree, run the scene, collect output/errors, and iterate.

Core Workflow

  1. Build a QA inventory before changing anything.

- Combine the user's requirements, the behaviors you implement, and the claims you expect to make in the final response. - For each claim, decide what evidence will prove it: scene tree state, validation output, runtime output/errors, manual in-engine verification, or a saved screenshot if the project supports it. - Add at least 2 off-happy-path checks for fragile gameplay, scene state, or editor synchronization.

  1. Confirm the active project and current scene state.

- Prefer godot_connection_status, godot_editor_get_project_info, and godot_editor_get_scene_tree. - If no live session exists yet, decide whether to start with file-based edits or connect immediately.

  1. Inspect before editing.

- For files: use godot_read_scene, godot_list_scene_nodes, godot_read_script, godot_read_shader, or godot_read_resource. - For live scenes: use godot_editor_get_scene_tree and optionally godot_editor_select_node. - For file-based scene checks, prefer godot_list_scene_nodes before godot_read_scene. - Do not read the full .tscn with godot_read_scene during a live task unless the live tree cannot answer the question. Large scene dumps slow the feedback loop. - Do not read the full .tscn during a file task unless you need serialized details like connections, external resources, or exact property payloads that godot_list_scene_nodes cannot provide.

  1. Make the smallest meaningful change.

- Prefer incremental edits over large speculative rewrites. - Keep node names and paths stable when possible so follow-up changes remain cheap. - When the goal is fast feedback rather than final UI quality, prefer a temporary probe node over exploratory edits to production nodes.

  1. Refresh or save based on how the change was made.

- External file edits while Godot is open: godot_editor_refresh_filesystem. - Live scene edits you intend to keep: godot_editor_save_scene.

  1. Validate before running.

- Use godot_validate_scene on touched scenes. - Use godot_validate_script on touched scripts. - Use godot_help or docs tools if a class/property choice is still uncertain.

  1. Run and inspect.

- Use godot_editor_run_scene. - Inspect godot_editor_get_errors, godot_editor_get_output, and, when needed, godot_editor_get_log_file. - Use godot_runtime_* tools for gameplay input, runtime waits, and screenshot capture when they are available. - Do not call godot_runtime_status before a scene is running unless you are checking whether a previous run is still active. Start the scene first. - After a live node add, prefer godot_editor_modify_node or structured properties on godot_editor_add_node before reaching for godot_editor_execute_gdscript. - Use godot_editor_execute_gdscript only for editor-side diagnostics or setup. It is not a general substitute for runtime gameplay input or frame capture. - If a screenshot or runtime assertion fails, inspect the specific node path and its properties first. Only expand to broad log review if the node-level check does not explain the failure.

  1. Update the QA inventory if exploration reveals new states, controls, failure modes, or visible claims.
  2. Sign off only from explicit evidence, and note any remaining manual verification gaps.

Fast Probe Presets

Use this path when the user wants faster feedback, a runtime target, or proof of interaction rather than final polish.

  • Prefer a temporary Label, Button, or LineEdit probe under a known live parent instead of spending turns rediscovering common Control properties.
  • Use the copyable presets in references/fast-probe-presets.md for deterministic geometry and obvious text.
  • Default to top-left anchored controls with fixed position, fixed size, and loud text or placeholder values so runtime click and screenshot evidence are unambiguous.
  • Save and validate immediately after adding the probe, then run the scene and gather evidence.
  • Remove the probe, save again, and revalidate before signoff unless the user explicitly wants to keep it.
  • Do not call docs tools or godot_help for common probe controls unless the preset fails, the project has unusual container behavior, or the user needs final-quality UI behavior rather than a temporary target.

Start Or Reuse Live Session

Keep the same bridge connection alive across iterations when possible.

Typical attach sequence:

godot_connect {}
godot_connection_status {}
godot_editor_get_project_info {}
godot_editor_get_scene_tree {}

If you need to override the configured bridge target for troubleshooting or multiple editors:

godot_connect { "host": "127.0.0.1", "port": 6550 }

If you need a specific scene:

godot_editor_open_scene { "scenePath": "res://scenes/main.tscn" }
godot_editor_get_scene_tree {}

Rules:

  • Reuse the existing connection instead of reconnecting on every turn.
  • If Godot was restarted, the project changed, or the bridge timed out, reconnect and reacquire the scene tree before making assumptions.
  • If a node path changed, rerun godot_editor_get_scene_tree instead of guessing the new path.

File-Based Workflow

Use file-based tools when the bridge is unavailable or when broad edits are easier outside the editor.

Common loop:

  1. Read the current artifact.
  2. Edit with the appropriate godot_* file tool.
  3. Validate the changed file.
  4. If Godot is open, call godot_editor_refresh_filesystem.
  5. Open or rerun the scene in the editor to verify behavior.

Typical pairings:

  • Scene structure: godot_read_scene, godot_add_node, godot_modify_node, godot_validate_scene
  • Scripts: godot_read_script, godot_write_script, godot_validate_script
  • Shaders: godot_read_shader, godot_write_shader
  • Resources: godot_read_resource, godot_write_resource
  • UI generation: godot_ui_* tools followed by validation and a live run pass

Use file reads only when you need file-only structure such as external resources, connections, or exact serialized properties. For most live scene checks, godot_editor_get_scene_tree is the faster baseline.

Refresh And Rerun Decision

  • External .tscn, .gd, .gdshader, .tres, or project.godot changes while the editor is open: run godot_editor_refresh_filesystem.
  • Live scene changes you want persisted: run godot_editor_save_scene.
  • Runtime behavior changes, script changes, or scene logic changes: rerun the scene.
  • Bridge disconnect, editor restart, or project reopen: reconnect and reacquire state.
  • Stale node paths: reread the scene tree before issuing more live node commands.

Runtime Automation Workflow

Use this whenever the current MCP surface exposes godot_runtime_* tools. If those tools are absent, only use it when a real project-side runtime harness already exists.

  • First inspect the available automation surface.

- Check the MCP tool list for dedicated godot_runtime_* tools first. - Search the project for an existing debug/test harness, local automation server, screenshot helper, or deterministic test runner. - Read the harness entrypoint before using it so you know whether it targets the editor process or the running game process.

  • Treat editor-side and runtime-side code as different environments.

- godot_editor_execute_gdscript runs in the editor/plugin context. - The current godot_runtime_* tools target the running game process through the AI Bridge runtime harness. - A project-specific fallback still needs its own harness, test scene, autoload, CLI, or transport to accept input commands and emit screenshots or state snapshots.

  • Prefer automation evidence in this order:

- dedicated godot_runtime_* tools - project-provided deterministic harness that writes artifacts or structured results - manual in-engine verification

  • For input automation:

- Prefer real input-event injection for UI, cursor, and focus-sensitive checks. - Prefer action-level helpers only for gameplay semantics where position and event routing do not matter. - Make the target control easy to hit before running: deterministic position, visible size, and obvious text or placeholder. - For temporary UI targets, start with the preset Label, Button, or LineEdit payloads instead of discovering properties from scratch.

  • For screenshot automation:

- Prefer deterministic output paths and stable capture timing. - Pair screenshots with structured state, logs, or scene assertions so visual claims are not based on a single image alone.

  • If no runtime harness exists, do not pretend the editor bridge can replace it. Call out the verification limit explicitly.

Checklists

Session Loop

  • Keep one live connection alive when the editor is available.
  • Reinspect current state before a new edit burst.
  • Make one coherent change set.
  • Refresh or save based on where the change happened.
  • Validate touched files.
  • Run the relevant scene.
  • Inspect runtime output and errors.
  • Update the QA inventory if new states or claims appear.

Validation And Runtime Checks

  • Structural changes: godot_validate_scene
  • Script changes: godot_validate_script
  • Runtime/editor issues: godot_editor_get_errors
  • General trace output: godot_editor_get_output
  • Longer or incremental log review: godot_editor_get_log_file
  • Focused introspection: godot_editor_execute_gdscript
  • Runtime automation: godot_runtime_status, godot_runtime_wait, godot_runtime_press_action, godot_runtime_release_action, godot_runtime_tap_action, godot_runtime_mouse_move, godot_runtime_click, godot_runtime_type_text, godot_runtime_capture_screenshot

When using godot_editor_execute_gdscript:

  • Keep scripts short and diagnostic.
  • Prefer returning structured values.
  • Treat the result as editor-side inspection evidence, not proof that a player-facing interaction works.

Signoff

  • Coverage is explicit against the QA inventory.
  • Touched scenes and scripts were validated.
  • The final scene tree or file structure was rechecked after edits.
  • A runtime pass was completed for behavior-affecting changes.
  • The response clearly distinguishes between validated behavior and assumptions.
  • If the claim depends on player input, timing, or visuals, manual in-engine verification or an existing automated harness was used.
  • Any missing evidence is called out explicitly.

Visual Evidence And Limits

This is still the largest difference from playwright-interactive, even after adding runtime tools.

  • godot-mcp now exposes first-class runtime input and screenshot tools through godot_runtime_*, but the workflow is still less browser-like than Playwright.
  • The bridge remains split across editor-side control and runtime-side automation. Keep those environments conceptually separate.
  • Do not claim visual QA is complete from scene tree and log output alone.
  • If a task depends on visuals, layout, animation timing, or gameplay feel, use godot_runtime_capture_screenshot or an equivalent project artifact when possible; otherwise inspect the running Godot window manually.
  • If the project can save screenshots to disk, review them directly and reference the path.
  • If no visual artifact exists, describe the verification limit plainly.

Common Failure Modes

  • godot_connect times out: Godot is not running, the AI Bridge plugin is disabled, or the port/host is wrong.
  • Not connected to Godot editor: reconnect before using godot_editor_* tools.
  • The shell workspace and the connected editor project differ: if the user asked to work against the running editor, continue on the connected project and mention the difference briefly instead of treating it as a blocker.
  • No scene open: open a scene explicitly with godot_editor_open_scene.
  • Node not found: reacquire the scene tree and verify the path before issuing another command.
  • External edits do not appear in the editor: run godot_editor_refresh_filesystem.
  • godot_editor_execute_gdscript compile failures: inspect godot_editor_get_errors and godot_editor_get_log_file.
  • Runtime automation tools time out or fail: make sure the scene is actually running, the runtime harness autoload is loaded, and the request targets the running game rather than the editor.
  • Older Godot or Claude-oriented notes may mention tools that are not in the current MCP surface. Use the reference file below and godot_help as the source of truth.

References

Read references/live-editor-tool-map.md when you need the current tool names, common payload shapes, or drift notes from the older Godot skill repo. Read references/fast-probe-presets.md when you need a fast temporary Label, Button, or LineEdit target for runtime or screenshot verification. Read references/runtime-automation-extension-points.md when you need the current limits, the recommended runtime harness shape, or the future runtime tool set this skill should prefer when it exists.

平台分布

Codex

36.33%
按下载量换算26

Claude

30.96%
按下载量换算22

Cursor

17.36%
按下载量换算12

Gemini CLI

9.71%
按下载量换算7

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

操作浏览器

该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

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。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills