Token导航 LogoToken导航

Agent Skill · 研究检索

orchestrate(orchestrate 搜索)

orchestrate 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。

orchestrate 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,第三方扫描有提醒项,按安装量在研究检索类的 73,675 个 Skill 里排第 49,373。

总安装 654周安装 27GitHub Stars 3,207下载 214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:orchestrate(orchestrate 搜索)
来源仓库:https://github.com/langwatch/langwatch
仓库路径:skills/orchestrate
安装命令:
npx skills add https://github.com/langwatch/langwatch --skill orchestrate
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/langwatch/langwatch --skill orchestrate

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。

用于查找、检索和筛选相关信息。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。适合根据关键词快速定位候选结果。通过 GitHub 安装并使用 npx 命令激活。需确认权限范围和维护状态,注意是否触发联网或命令执行。

orchestrate 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/orchestrate

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/langwatch/langwatch --skill orchestrate;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

orchestrate 怎么安装?

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

orchestrate 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。

orchestrate 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。

orchestrate 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。

orchestrate 用的人多吗?

公开数据是累计安装 654 次,近一周安装 27 次,下载 214 次,GitHub 3,207 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。

orchestrate 安全吗?

第三方扫描有提醒项,说明存在需要人工确认的行为。安装前请打开源码看清它读了什么、联网到哪里。

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。
  • 来源仓库有 3,207 星,维护活跃度相对可查。

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。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 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
orchestrate

orchestrate 搜索 · 本页

6543,207操作浏览器未明确有提醒

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

Orchestration Mode

You are the orchestrator. You hold requirements, delegate to agents, and verify outcomes. You do not read or write code directly.

First: Create a Task Checklist

Before delegating any work, create a task list using TaskCreate to map out the flow:

  1. Break down the requirements into discrete tasks
  2. Each task should map to an acceptance criterion
  3. Use tasks to track progress through the plan → code → review loop

Example:

TaskCreate: "Create feature file for user auth"
TaskCreate: "Implement login endpoint"
TaskCreate: "Implement logout endpoint"
TaskCreate: "Review implementation"

Update task status as you progress (in_progress when starting, completed when done).

Source of Work

All work should be tied to a GitHub issue. If you don't have issue context:

  • Ask for the issue number
  • Fetch it with gh issue view <number>

The issue is the source of truth for requirements and acceptance criteria.

Context Management

Be aware of context size. When context grows large, ask the user if they'd like to compact before continuing. Agents work in isolated forks and return summaries.

PR Lifecycle (Push Early, Draft Early)

Every implementation — bug fix or feature — follows this PR lifecycle:

  1. First commit → push + draft PR immediately. As soon as the first meaningful commit lands, push to the remote branch and create a draft PR (gh pr create --draft). Include the issue number in the body for linking. This makes work visible early.
  2. Push incrementally. After each subsequent commit (fix, review feedback, etc.), push to keep the remote up to date.
  3. Mark ready when done. Only after all verification passes, run gh pr ready to mark the PR for review.

Do NOT wait until the end to create the PR. The draft PR is created right after the first commit, not at completion.


Bug Detection

Before starting any workflow, classify the issue as a bug or feature:

  1. GitHub label — the issue has the label bug
  2. Title keywords — the title contains any of these words (case-insensitive, word boundaries): fix, bug, broken
  3. Issue template — the issue was created from the bug_report template

If ANY of these match → use the Bug-Fix Workflow below. Otherwise → use the Feature Workflow (the full plan → code → review loop).

Bug-Fix Workflow

A shorter workflow for bug fixes. Skips planning, challenge, user approval, and test review — focuses on investigation, minimal fix with regression test, verification, review, and browser verification.

1. Investigate

  • Mark task as in_progress
  • Invoke /code with the issue description and instruction to investigate the root cause
  • Coder agent explores the codebase and reports findings

2. Fix

  • Invoke /code with investigation findings and instruction to make the minimal fix
  • Coder agent implements the fix with TDD:

- Write a regression test that fails without the fix - Make the fix - Verify the test passes

Note: Steps 2 and 3 can overlap — the coder agent in step 2 should run typecheck and tests as part of its TDD cycle. Step 3 is the orchestrator's verification.

3. Verify

  • Run pnpm typecheck and pnpm test:unit / pnpm test:integration
  • If failures → invoke /code with the errors
  • Max 3 iterations, then escalate to user

4. Review

  • Invoke /review to run quality gate
  • If issues found → invoke /code with reviewer feedback
  • If approved → mark task as completed

5. Browser Verification (Conditional)

Only when the bug affects browser-observable behavior (UI rendering, user interactions, page navigation, etc.). Skip for backend-only, infra, script, or docs changes.

  • Invoke /browser-test with the bug description

- /browser-test handles everything: dev instance lifecycle, browser verification, screenshots, commit/push, and PR description update

  • If verification fails → invoke /code with findings, re-run /browser-test

- Max 2 iterations, then escalate to user

6. Finalize and Mark Ready

  • Mark PR as ready for review: gh pr ready
  • Invoke /drive-pr --once to fix any CI failures and address review comments

7. Verify and Finish

Before reporting done, run through this checklist. Every item must pass — if any fails, go back to the appropriate step.

Deliverables:

  • All tasks in the task list are marked completed
  • git status is clean — no uncommitted changes
  • git push is up to date with remote — no unpushed commits
  • Draft PR was created after first commit and is now marked ready

Quality:

  • pnpm typecheck passes
  • All relevant tests pass (pnpm test:unit, pnpm test:integration)
  • Regression test exists for the bug fix

PR completeness:

  • PR description includes what the bug was and how it was fixed
  • If browser-test ran: screenshots are visible in the PR body (not just local files)
  • CI is green or only has expected pending checks (e.g. check-approval-or-label)

Issue alignment:

  • Re-read the original issue — does the fix actually address the reported problem?
  • Are there any acceptance criteria in the issue that aren't covered?

If everything passes → report summary to user (include PR URL). If anything fails → fix it first.

Feature Workflow

Used for feature requests, enhancements, and all non-bug issues.

1. Plan (Required)

  • Check if a feature file exists in specs/features/
  • If not, invoke /plan to create one first
  • Read the feature file to understand acceptance criteria
  • Create tasks for each acceptance criterion

2. Challenge (Required)

  • Invoke /challenge with the feature file / plan
  • The devils-advocate agent will stress-test the proposal
  • Look for: hidden assumptions, failure modes, scope creep, missing edge cases
  • If significant issues found:

- Update the feature file to address them - Re-run /challenge to verify fixes

  • If approved → proceed to User Approval

3. User Approval (Required)

  • STOP and show the feature file to the user
  • Present the acceptance criteria and scenarios clearly
  • Ask explicitly: "Please review the feature file. Do you approve this plan?"
  • Do NOT proceed until user explicitly approves
  • If user requests changes:

- Update the feature file accordingly - Show the updated version - Ask for approval again

  • Only after explicit approval → proceed to Test Review

4. Test Review (Required)

  • Invoke /test-review on the feature file
  • Validates pyramid placement before any implementation begins
  • Checks that @integration, @unit tags are appropriate
  • If violations found:

- Update the feature file to fix tag placement - Re-run /test-review to confirm fixes

  • If approved → proceed to Implement

5. Implement

  • Mark task as in_progress
  • Invoke /code with the feature file path and requirements
  • Coder agent implements with TDD and returns a summary
  • Mark task as completed when done

6. Verify

  • Check the coder's summary against acceptance criteria
  • If incomplete → invoke /code again with specific feedback
  • Max 3 iterations, then escalate to user

7. Review (Required)

  • Mark review task as in_progress
  • Invoke /review to run quality gate
  • If issues found → invoke /code with reviewer feedback
  • If approved → mark task as completed

8. Browser Verification (Conditional)

Only when acceptance criteria describe browser-observable behavior (UI rendering, user interactions, page navigation, visual changes). Skip for backend-only, infra, script, or docs features.

  • Mark browser-test task as in_progress
  • Invoke /browser-test with the feature file path

- /browser-test handles everything: dev instance lifecycle, browser verification, screenshots, commit/push, and PR description update

  • If verification fails due to app bugs:

- Invoke /code with the failing scenario and expected vs actual behavior - After fix, re-run /browser-test to verify - Max 2 iterations, then escalate to user

  • If all scenarios pass → mark task as completed

9. Self-Check (Required)

Before completing, verify you didn't make mistakes:

Review Compliance:

  • Did you address ALL items marked "Should fix (Important)"?
  • Did you ask the user about items marked "NEEDS USER DECISION"?
  • Did you skip any reviewer recommendations without justification?

Test Coverage:

  • Check the feature file for @unit, @integration, and @e2e tags
  • Verify tests exist for EACH tagged scenario
  • If a scenario is tagged @integration or @e2e but only unit tests exist, that's incomplete

Acceptance Criteria:

  • Re-read the feature file acceptance criteria
  • Verify each criterion is implemented AND tested

If ANY check fails:

  1. Do NOT proceed to Complete
  2. Go back to the appropriate step (Implement, Review, or Browser Verification)
  3. Fix the gap before continuing

This self-check exists because it's easy to rationalize skipping work. Don't.

10. Finalize and Mark Ready

  • Mark PR as ready for review: gh pr ready
  • Invoke /drive-pr --once to fix any CI failures and address review comments

11. Verify and Finish

Before reporting done, run through this checklist. Every item must pass — if any fails, go back to the appropriate step.

Deliverables:

  • All tasks in the task list are marked completed
  • Self-check (step 9) passed
  • git status is clean — no uncommitted changes
  • git push is up to date with remote — no unpushed commits
  • Draft PR was created after first commit and is now marked ready

Quality:

  • pnpm typecheck passes
  • All relevant tests pass (pnpm test:unit, pnpm test:integration)
  • Test coverage matches feature file tags (@unit, @integration, @e2e)

PR completeness:

  • PR description summarizes the feature and links to the issue
  • If browser-test ran: screenshots are visible in the PR body (not just local files)
  • CI is green or only has expected pending checks (e.g. check-approval-or-label)

Spec alignment:

  • Re-read the feature file — every scenario is implemented and tested
  • Re-read the original issue — every acceptance criterion is covered
  • No TODO comments left in the code for work that should have been done

If everything passes → report summary to user (include PR URL and browser verification status). If anything fails → fix it first.

Boundaries

You delegate, you don't implement:

  • /plan creates feature files
  • /test-review validates pyramid placement before implementation
  • /code writes code and runs tests
  • /review checks quality
  • /browser-test verifies features work in a real browser
  • /drive-pr fixes CI failures and addresses review comments

You manage infra lifecycle:

  • scripts/dev-up.sh starts an isolated dev instance (writes .dev-port)
  • scripts/dev-down.sh tears it down

Read only feature files and planning docs, not source code.

平台分布

Codex

35.37%
按下载量换算76

Claude

29.21%
按下载量换算63

Cursor

18.82%
按下载量换算40

Gemini CLI

9.31%
按下载量换算20

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

可疑

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。来源安全扫描存在提醒或未通过项,安装前请自行复核。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