Token导航 LogoToken导航

Agent Skill · 研究检索

plugin-workflow(插件工作流程)

plugin-workflow 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。

plugin-workflow 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,权限为接触敏感数据;许可证为 MIT,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 60,316。

总安装 371周安装 15GitHub Stars 182下载 116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
MIT

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:plugin-workflow(插件工作流程)
来源仓库:https://github.com/glittercowboy/plugin-freedom-system
仓库路径:skills/plugin-workflow
安装命令:
npx skills add https://github.com/glittercowboy/plugin-freedom-system --skill plugin-workflow
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。不同来源提供的安装方式可能略有差异;本站展示可直接复制的安装命令,安装前请核对来源页面。

skills.shnpx skills
npx skills add https://github.com/glittercowboy/plugin-freedom-system --skill plugin-workflow

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,支持关键词匹配与任务场景适配。适合在 Codex、Claude、Cursor、Gemini CLI 中快速定位候选结果。可通过 npx skills add 命令从指定仓库安装并使用。安装前建议确认权限范围和维护状态,避免触发联网或文件读写操作。

注意:实际功能需结合原始 README 进一步核验,当前描述基于通用模板。

来源仓库内路径:skills/plugin-workflow

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/glittercowboy/plugin-freedom-system --skill pl…;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

    它需要接触敏感数据,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

plugin-workflow 怎么安装?

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

plugin-workflow 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。

plugin-workflow 需要什么权限?

权限级别是接触敏感数据。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。

plugin-workflow 是免费的吗?

来源标注的许可证是 MIT。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

plugin-workflow 用的人多吗?

公开数据是累计安装 371 次,近一周安装 15 次,下载 116 次,GitHub 182 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。

plugin-workflow 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。
  • 3 个目录同时收录,说明这个 Skill 流传比较广。
  • 许可证明确为 MIT,商用和分发边界好判断。

先想清楚的地方

  • 需要接触敏感数据,装到常用环境前要先评估影响面。
  • 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

  • 环境里有大量密钥、凭证,不方便给第三方 Skill 读取的人。
  • 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
plugin-workflow

插件工作流程 · 本页

371182接触敏感数据MIT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

plugin-workflow Skill

Purpose: Pure orchestrator for stages 1-3 of JUCE plugin implementation with automatic validation after each stage. This skill delegates to specialized subagents and validation-agent for continuous quality assurance.

Overview

Implementation milestones:

  • Build System Ready (Stage 1): Create build system and implement parameters (foundation-shell-agent)
  • Audio Engine Working (Stage 2): Implement audio processing (dsp-agent)
  • UI Integrated (Stage 3): Connect WebView interface to audio engine (gui-agent)

Stage 0 (Research & Planning) is handled by plugin-planning skill.

After Stage 3 completes, plugin is ready for installation (no separate validation stage - validation is automatic and continuous).

Delegation Protocol

CRITICAL: Stages 1-3 MUST invoke subagents via Task tool. This skill is a pure orchestrator and NEVER implements plugin code directly.

Delegation sequence for every stage:

  1. Load contracts in parallel (architecture.md, plan.md, parameter-spec.md, creative-brief.md)
  2. Read Required Reading (juce8-critical-patterns.md) once at workflow start
  3. Construct minimal prompt with plugin name + stage + Required Reading
  4. Invoke subagent via Task tool
  5. After subagent returns, invoke validation-agent (ALL stages 1-3)
  6. Execute checkpoint protocol (see references/checkpoint-protocol.md)

Stage routing:

  • Stage 1 → foundation-shell-agent
  • Stage 2 → dsp-agent
  • Stage 3 → gui-agent

Validation routing: After each stage completes, validation-agent runs automatically with enhanced runtime validation (compile-time + runtime tests). If validation fails with continue_to_next_stage: false, workflow BLOCKS until issues resolved.

Preconditions

Before starting Stage 1, verify these contract files exist:

  • plugins/$PLUGIN_NAME/.ideas/architecture.md (from Stage 0)
  • plugins/$PLUGIN_NAME/.ideas/plan.md (from Stage 0)
  • plugins/$PLUGIN_NAME/.ideas/creative-brief.md (from ideation)
  • plugins/$PLUGIN_NAME/.ideas/parameter-spec.md (from UI mockup finalization)

If parameter-spec-draft.md exists but parameter-spec.md missing: Block with message: "Draft parameters found, but full specification required. Complete UI mockup workflow to generate parameter-spec.md. Run: /dream [PluginName] → option 2 (Full UI mockup first)"

If contracts missing: Block and instruct user to run /plan [PluginName] to complete Stage 0.

See references/precondition-checks.md for implementation.

Resume Entry Point

When resuming via /continue [PluginName]:

  1. Verify state integrity (see references/state-management.md#verifyStateIntegrity)
  2. Parse .continue-here.md for current stage and workflow mode
  3. Verify contracts unchanged since last checkpoint (checksums match)
  4. Verify git working directory clean
  5. Verify PLUGINS.md status matches.continue-here.md stage

If all checks pass: Resume at stage specified in.continue-here.md If any fail: Present recovery menu (reconcile / clean working directory / review changes)

Workflow Mode

Determine whether to auto-progress (express mode) or present menus (manual mode).

Mode sources (priority order):

  1. Environment variables: WORKFLOW_MODE=express|manual
  2. .continue-here.md field (for resumed workflows)
  3. Default to "manual"

Express mode behavior:

  • Auto-progress through stages without menus
  • Drops to manual on ANY error (build failures, validation failures, etc.)
  • Final menu always appears after Stage 3 (plugin complete)

See references/workflow-mode.md for implementation.

Stage Dispatcher

Entry point: Called by /implement or /continue after plugin-planning completes.

Dispatch flow:

  1. Verify state integrity → BLOCK if corrupted (exit 2 → run /reconcile)
  2. Check preconditions → BLOCK if failed
  3. Automatic brief sync (before Stage 1 only, if mockup exists) → See references/creative-brief-sync.md
  4. Route to subagent based on stage
  5. Pass contracts and Required Reading to subagent
  6. Wait for subagent completion
  7. Invoke validation-agent with enhanced runtime validation
  8. Execute checkpoint protocol

See references/dispatcher-pattern.md for full algorithm.

Phase-Aware Dispatch

For Stages 2-3 with complexity ≥3, use phase-aware dispatch to incrementally implement complex plugins.

When to use:

  • Stage 2 (DSP) or Stage 3 (GUI)
  • Complexity score ≥3 (from plan.md)
  • plan.md contains phase markers (### Phase 2.X or ### Phase 3.X)

How it works:

  1. Detect phases by scanning plan.md for phase markers
  2. Loop through phases sequentially (Phase 2.1 → 2.2 → 2.3...)
  3. Invoke subagent once per phase with phase-specific prompt
  4. Run validation-agent after each phase
  5. Execute checkpoint protocol after each phase
  6. Present decision menu showing progress ("Phase 2 of 4 complete")

CRITICAL: Never send "Implement ALL phases" to subagent. This caused DrumRoulette Stage 2 compilation errors. Phase-aware dispatch is MANDATORY for complex plugins.

See references/phase-aware-dispatch.md for detailed algorithm.

Checkpoint Protocol

After EVERY subagent return, execute this 6-step sequence:

  1. Verify state update: Check subagent updated.continue-here.md and PLUGINS.md
  2. Fallback state update: If verification fails, orchestrator updates state
  3. Invoke validation: Run validation-agent for ALL stages 1-3 (BLOCKING on runtime failures)
  4. Commit stage: Auto-commit all changes with git
  5. Verify checkpoint: Validate all steps completed successfully
  6. Handle checkpoint: Present menu (manual mode) or auto-progress (express mode)

Checkpoint applies to:

  • Simple plugins (complexity ≤2): After stages 1, 2, 3
  • Complex plugins (complexity ≥3): After stage 1 AND after EACH DSP/GUI phase (2.X, 3.X)

See references/checkpoint-protocol.md for implementation.

Validation Integration

Stages 1-3 invoke validation-agent with enhanced runtime validation after subagent completes:

  • BLOCKING on runtime failures: If status: FAIL and continue_to_next_stage: false, workflow stops
  • Runs compile-time checks (contract matching, implementation correctness)
  • Runs runtime tests (load plugin, process audio, parameter changes) when binary available
  • Returns JSON report with status, checks, recommendation
  • Max 500 tokens per report

Blocking behavior:

  • If validation passes (PASS/WARNING): Continue to next stage
  • If validation fails with continue_to_next_stage: true: Present warning, allow continuation
  • If validation fails with continue_to_next_stage: false: BLOCK workflow, present error menu

See references/validation-integration.md for functions.

Subagent Handoff Protocol

Subagents update state files AND return JSON report:

{
  "status": "success" | "error",
  "stage": 1-3,
  "completionStatement": "...",
  "filesCreated": [...],
  "nextSteps": [...],
  "stateUpdated": true | false,
  "stateUpdateError": "..." (optional)
}

Verification:

  1. Check stateUpdated field in JSON report
  2. If true: Verify.continue-here.md actually changed
  3. If false/missing: Trigger orchestrator fallback

Fallback: Orchestrator reads current state, updates fields, writes back.

See references/state-management.md for fallback implementation.

Required Reading Injection

All subagents (stages 1-3) receive troubleshooting/patterns/juce8-critical-patterns.md to prevent repeat mistakes.

Implementation:

  1. Read juce8-critical-patterns.md ONCE at workflow start
  2. Prepend to all subagent prompts with clear separator
  3. Pass to each subagent invocation from memory (no re-reading)

Reference Files

Each stage has detailed documentation in references/:

Integration Points

Invoked by:

  • /implement command (after plugin-planning completes)
  • /continue command (for stages 1-3)
  • context-resume skill (when resuming implementation)

Invokes via Task tool:

  • foundation-shell-agent (Stage 1) - REQUIRED
  • dsp-agent (Stage 2) - REQUIRED
  • gui-agent (Stage 3) - REQUIRED
  • validation-agent (Stages 1-3) - REQUIRED, BLOCKING on runtime failures

Also invokes:

  • build-automation skill (build verification)
  • plugin-lifecycle skill (if user chooses to install)

Reads (contracts):

  • architecture.md, plan.md, creative-brief.md, parameter-spec.md

Creates:

  • .continue-here.md (handoff file)

Updates:

  • PLUGINS.md (status after each stage)
  • .continue-here.md (after each stage)

Deletes after Stage 3:

  • .continue-here.md (workflow complete, plugin ready for installation)

Error Handling

Contract files missing before Stage 1: Block and instruct user to run /plan [PluginName].

Build fails during subagent execution: Subagent returns error. Present menu:

  1. Investigate (deep-research)
  2. Show code
  3. Show build output
  4. Manual fix (resume with /continue)

State mismatch detected (exit 2): BLOCKING error - user must run /reconcile [PluginName] to fix.

Validation fails with continue_to_next_stage: false: BLOCKING error. Present menu with investigation options. Workflow cannot proceed until issues resolved.

Validation fails with continue_to_next_stage: true: Present warning, allow user to decide whether to continue or fix issues first.

See references/error-handling.md for detailed patterns.

Decision Menu Protocol

Use inline numbered menus for:

  • After EVERY stage completion (checkpoint gates)
  • Build failure recovery
  • Test failure investigation
  • Phase completion (for complex plugins)

Format:

✓ [Milestone name]

What's next?

1. [Next milestone action] (recommended)
2. [Run tests] - Verify implementation
3. [Pause workflow] - Resume anytime
4. [Review code] - See what was implemented
5. Other

Choose (1-5): _

Express mode: Skip menus and auto-progress to next stage (except final stage). Manual mode: ALWAYS wait for user response.

Success Criteria

Workflow succeeds when:

  • All subagents (stages 1-3) invoked successfully via Task tool
  • Plugin compiles without errors at each stage
  • All validation passes (or explicitly allowed to continue with warnings)
  • All stages completed in sequence (1 → 2 → 3)
  • Decision menus presented after EVERY stage (manual mode)
  • PLUGINS.md updated to ✅ Working after Stage 3
  • Handoff file deleted after Stage 3 (workflow complete)
  • Git history shows atomic commits for each stage

Anti-Patterns

Common pitfalls to AVOID:

CRITICAL:

  • ❌ Implementing stage logic directly in orchestrator
  • ✓ ALWAYS use Task tool to invoke appropriate subagent

CRITICAL:

  • ❌ Sending "Implement ALL phases" to subagent for Stages 2-3
  • ✓ ALWAYS detect phases in plan.md and loop through them one at a time

CRITICAL:

  • ❌ Proceeding to next stage when validation fails with continue_to_next_stage: false
  • ✓ BLOCK workflow and present error menu until issues resolved

HIGH:

  • ❌ Not verifying subagent updated state
  • ✓ Check stateUpdated field, verify.continue-here.md changed, fallback if needed

HIGH:

  • ❌ Skipping phase detection for Stages 2-3 when complexity ≥3
  • ✓ Read plan.md to check for phases BEFORE invoking dsp-agent or gui-agent

HIGH:

  • ❌ Skipping validation after subagent completes
  • ✓ ALWAYS invoke validation-agent after each stage (1-3)

MEDIUM:

  • ❌ Not injecting Required Reading to subagents
  • ✓ Always pass juce8-critical-patterns.md to prevent repeat mistakes

平台分布

Claude Code

31.37%
按下载量换算36

windsurf

23.18%
按下载量换算27

OpenCode

18.61%
按下载量换算22

Codex

13.64%
按下载量换算16

Antigravity

7.69%
按下载量换算9

Gemini CLI

3.58%
按下载量换算4

安全审计

Gen Agent Trust Hub

通过

Socket

可疑

Snyk

通过

权限和风险

接触敏感数据

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。来源安全扫描存在提醒或未通过项,安装前请自行复核。

继续浏览同类 Skills