agentsclimarketplace

Multi agent workflow designer

Skill findscripter/everything-skills/04-ai/multi-agent-workflow-designer

类书式 AI Agent 技能大典 · 精选/中文化/互见成网的 500+ 开源技能,可作为 Claude Code 插件市场一键安装。A curated, cross-referenced encyclopedia of 500+ open-source agent skills.

Install
npx -y skills add findscripter/everything-skills --skill multi-agent-workflow-designer

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

当设计多步骤 Agent 流水线、在单 Agent 与多 Agent 之间抉择、或重构受上下文膨胀/交接不可靠困扰的 LLM 工作流时使用;做选定编排模式(顺序/并行/路由/编排器/评估器)、用脚手架脚本生成 JSON 骨架配置、定义交接契约与重试/超时/预算护栏,产出可落地的工作流蓝图;不适用于单条良构提示词即可解决的简单任务。触发词:多智能体工作流、Agent 编排、交接契约

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

5.6 KB, ~1.6k tokens by cl100k_base, as published. Nobody here has run it

何时使用

当一个任务复杂到单条提示词无法稳定胜任,需要把它拆成多个有明确边界的专职 Agent 协作时使用本技能。典型场景:

  • 一条提示词不足以承载任务复杂度,需要带显式边界的专家 Agent。
  • 希望在动手实现前先定下确定性的工作流结构。
  • 需要质量/安全关卡的校验回路。
  • 正在重构一个受上下文膨胀或交接不可靠困扰的 LLM 工作流。
  • 在「单 Agent vs 多 Agent」之间做架构抉择。

不该用的边界:若任务用一条良构的提示词即可解决,不要为它套多 Agent 编排——过度编排会徒增成本、延迟与故障面。先用能满足需求的最小模式。

步骤

  1. 选模式:依据「依赖形态 + 风险画像」从五种模式中选型(见下方模式表)。
  2. 生成骨架:用 scripts/workflow_scaffolder.py 生成 JSON 骨架配置。
  3. 定义交接契约:为每一条边(edge)补齐交接契约字段。
  4. 加护栏:为每一步加上重试/超时与输出校验关卡。
  5. 小预算试跑:在放大规模前,用较小的上下文预算做 dry-run 验证。

指令

模式选型表(依赖形态 → 模式):

  • sequential:严格逐步依赖链——每一步都依赖上一步输出。
  • parallel:独立子任务先扇出(fan-out)后扇入(fan-in),用于提吞吐、降延迟。
  • router:按意图/类型分派到专职处理器,并配 fallback 兜底。
  • orchestrator:规划者(planner)协调多专家,按依赖(DAG)动态规划。
  • evaluator:生成器 + 质量关卡回路,正确性/质量必须把关时用。

生成骨架命令:

# 生成顺序工作流骨架
python3 scripts/workflow_scaffolder.py sequential --name content-pipeline

# 生成编排器工作流并落盘保存
python3 scripts/workflow_scaffolder.py orchestrator --name incident-triage --output workflows/incident-triage.json

脚本支持的 pattern 取值:sequential | parallel | router | orchestrator | evaluator;可选参数 --name(默认 new-workflow)、--output(落盘路径,缺省则打印到 stdout)。

交接最小契约(每条边都必须携带这些字段):

  • workflow_id
  • step_id
  • task
  • constraints
  • upstream_artifacts(只传定向产物,不要整段上游上下文)
  • budget_tokens
  • timeout_seconds

示例

顺序模式骨架(含内建重试):

{
  "name": "content-pipeline",
  "pattern": "sequential",
  "steps": [
    {"id": "research", "agent": "researcher", "next": "draft"},
    {"id": "draft", "agent": "writer", "next": "review"},
    {"id": "review", "agent": "reviewer", "next": null}
  ],
  "retry": {"max_attempts": 2, "backoff_seconds": 2}
}

评估器模式骨架(质量回路):

{
  "pattern": "evaluator",
  "generator": {"agent": "generator"},
  "evaluator": {"agent": "evaluator", "criteria": ["accuracy", "format", "safety"]},
  "loop": {"max_iterations": 3, "pass_threshold": 0.8, "on_fail": "revise_and_retry"}
}

编排器模式骨架(DAG 依赖 + 受控并行):

{
  "pattern": "orchestrator",
  "orchestrator": {"agent": "orchestrator", "planning": "dynamic"},
  "specialists": ["researcher", "coder", "analyst", "writer"],
  "execution": {"dependency_mode": "dag", "max_parallel": 3, "completion_policy": "all_required"}
}

注意事项

常见陷阱:

  • 对一条良构提示词就能解决的任务过度编排。
  • 给外部模型调用漏配超时/重试策略。
  • 把整段上游上下文一股脑往下传,而非只传定向产物(artifacts)。
  • 忽视逐步累积的单步成本。

最佳实践:

  1. 从能满足需求的最小模式起步。
  2. 交接载荷保持显式且有界。
  3. 扇入(fan-in)综合前,先校验各路中间产物。
  4. 每一步都强制预算与超时上限。

互见

  • 模式模板详表:源技能 references/workflow-patterns.md(顺序/并行/路由/编排器/评估器各模式的 JSON 模板与选型启发式)。
  • 脚手架脚本:scripts/workflow_scaffolder.py
  • 智能/agents 域内其他 Agent 编排、提示词工程类技能。

采编自 alirezarezvani/claude-skills(MIT),适配重写为中文版,非逐字翻译。

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.