swarm-meeting
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseUse this skill to turn a vague "let's discuss" request into a bounded decision meeting.
使用该技能将模糊的“我们来讨论一下”请求转化为一场有明确边界的决策会议。
Inputs
输入信息
Provide:
- topic
- need
- profile name
- optional artifacts, constraints, decision owner, and timebox
If the request spans many files or docs, prepare a brief pack first with or by reading the smallest relevant local artifacts directly.
swarm-research-pack请提供:
- topic
- need
- profile name
- 可选的工件、约束条件、决策负责人和timebox
如果请求涉及多个文件或文档,请先使用准备一份简短的资料包,或直接阅读最相关的小型本地工件。
swarm-research-packAmbiguous Input Assist
模糊输入辅助
When the user asks for a meeting without enough structure, do not stop immediately. Convert the vague request into a short guided start.
- Infer as the default profile unless the user clearly asks for stronger conflict or a deeper architecture review.
balanced - Infer a default of
needwhen the user does not specify one.options, objections, and recommendation - Rephrase the likely topic from the active conversation or nearest artifact.
- Show a short kickoff block before the meeting if key inputs are missing.
Use a compact pattern like this:
markdown
I can run this as a `balanced` swarm meeting.
- Topic: <inferred topic or "please provide one sentence">
- Need: options, objections, and recommendation
- Other profiles: `architecture-debate`, `extreme-diversity`
Examples:
- `Use the swarm-meeting skill. Topic: ... Need: ...`
- `Use swarm-meeting with profile: architecture-debate ...`Keep this assist block short. Do not turn it into a questionnaire unless the missing information would materially change the outcome.
当用户发起的会议请求结构不够清晰时,不要立即终止。将模糊请求转化为简短的引导式开场。
- 除非用户明确要求更强的冲突或更深入的架构评审,否则默认使用配置文件。
balanced - 当用户未指定需求时,默认需求为(方案、异议与建议)。
options, objections, and recommendation - 从当前对话或最近的工件中推断可能的主题。
- 如果缺少关键输入,在会议前展示一个简短的启动模块。
使用如下简洁格式:
markdown
I can run this as a `balanced` swarm meeting.
- Topic: <inferred topic or "please provide one sentence">
- Need: options, objections, and recommendation
- Other profiles: `architecture-debate`, `extreme-diversity`
Examples:
- `Use the swarm-meeting skill. Topic: ... Need: ...`
- `Use swarm-meeting with profile: architecture-debate ...`保持该辅助模块简短。除非缺失的信息会实质性影响结果,否则不要将其变成问卷调查。
Workflow
工作流程
- Load exactly one profile from .
profiles/ - Load only the role prompts needed for that profile from .
prompts/ - Rephrase the decision in 1-2 sentences and define the agenda.
- Collect independent role memos before synthesis. Keep each memo short: 3-7 bullets or short paragraphs.
- Run the objection round if the profile requires it.
- Force a recommendation, open questions, and concrete next actions.
- Separate facts from inferences. Mark assumptions explicitly.
- 从中加载恰好一个配置文件。
profiles/ - 从中仅加载该配置文件所需的角色提示词。
prompts/ - 用1-2句话重新表述决策内容并确定议程。
- 在综合汇总前收集独立的角色备忘录。每份备忘录要简短:3-7个要点或短段落。
- 如果配置文件要求,开展异议环节。
- 明确给出建议、待解决问题和具体的后续行动。
- 区分事实与推断。明确标记假设内容。
Guardrails
约束规则
- Treat roles as evaluation biases, not personalities.
- Keep between 4 and 6 unless the user explicitly asks for a larger meeting.
max_agents - Prefer local role emulation by default. Only spawn subagents when the user explicitly asks for delegation or parallel agents, or when the environment clearly allows it.
- Do not let any single role dominate the summary.
- Avoid unbounded brainstorming. End with a decision or a clearly framed fork.
- If the user input is underspecified, provide the short assist block above, make one reasonable default assumption, and continue.
- 将角色视为评审视角偏差,而非个性特征。
- 除非用户明确要求更大规模的会议,否则保持在4到6之间。
max_agents - 默认优先采用本地角色模拟。仅当用户明确要求委托或并行Agent,或者环境明确允许时,才生成子Agent。
- 不要让任何单一角色主导总结内容。
- 避免无边界的头脑风暴。以决策或明确的分支选项结束会议。
- 如果用户输入信息不明确,提供上述简短的辅助模块,做出一个合理的默认假设并继续推进。
Output Contract
输出规范
Use this section order unless the user asks for a different format:
markdown
undefined除非用户要求其他格式,否则请遵循以下章节顺序:
markdown
undefinedDesign Review
Design Review
Problem Statement
Problem Statement
Agenda
Agenda
Role Memos
Role Memos
Options Considered
Options Considered
Recommended Architecture
Recommended Architecture
MVP Plan
MVP Plan
Risks and Failure Modes
Risks and Failure Modes
Open Questions
Open Questions
Concrete Next Actions
Concrete Next Actions
undefinedundefinedResources
资源
- : default 5-role review for most implementation decisions.
profiles/balanced.yaml - : stronger architecture pressure with semantics and platform concerns.
profiles/architecture-debate.yaml - : intentionally conflicting evaluation axes for high-stakes tradeoffs.
profiles/extreme-diversity.yaml - : short role cards. Load only selected roles.
prompts/*.md - : ready-to-copy invocation examples for common meeting intents.
references/prompt-examples.md - : explicit rules for when to stay local versus when to spawn subagents.
references/operating-rules.md - : migrated example from the original one-off prompt.
references/task-scoped-context-review.md - : recommend or inspect a profile without external dependencies.
scripts/select_profile.py - : render a meeting brief from a profile plus topic, need, and artifacts.
scripts/render_meeting_context.py
- : 适用于大多数实现决策的默认5角色评审配置。
profiles/balanced.yaml - : 针对语义与平台问题进行更严格的架构测试。
profiles/architecture-debate.yaml - : 针对高风险权衡场景设置刻意冲突的评审维度。
profiles/extreme-diversity.yaml - : 简短的角色卡片。仅加载选定的角色。
prompts/*.md - : 针对常见会议意图的可直接复制调用示例。
references/prompt-examples.md - : 明确说明何时采用本地模拟、何时生成子Agent的规则。
references/operating-rules.md - : 从原始一次性提示词迁移而来的示例。
references/task-scoped-context-review.md - : 无需外部依赖即可推荐或查看配置文件的脚本。
scripts/select_profile.py - : 根据配置文件、主题、需求和工件生成会议简报的脚本。
scripts/render_meeting_context.py
Role Selection Rules
角色选择规则
- Always include .
moderator - Include unless the user explicitly wants a fast, low-friction pass.
skeptic - Prefer when delivery pressure is real.
pragmatist - Prefer for MVP, operational simplicity, or "ship this week".
minimalist - Prefer for platform or long-horizon design.
maximalist - Prefer when the topic involves search, indexing, code selection, Tree-sitter, LSP, or semantics.
retrieval-architect - Prefer when the topic involves task hubs, handoff, checkpointing, CLI flow, or session lifecycle.
granary-workflow-lead - Prefer when correctness depends on type, symbol, or compiler information.
semantic-purist - Prefer when a simple
lightweight-contrarian/diff/docs index may beat a heavy platform.rg - Prefer when token budget, context pollution, or fresh-session strategy matters.
context-economist - Prefer when tool sprawl is the main risk.
workflow-conservative
- 始终包含(主持人)角色。
moderator - 除非用户明确要求快速、低摩擦的评审,否则包含(质疑者)角色。
skeptic - 当存在交付压力时,优先选择(实用主义者)角色。
pragmatist - 针对MVP、操作简洁性或“本周上线”需求,优先选择(极简主义者)角色。
minimalist - 针对平台或长期规划设计,优先选择(极致主义者)角色。
maximalist - 当主题涉及搜索、索引、代码选择、Tree-sitter、LSP或语义相关内容时,优先选择(检索架构师)角色。
retrieval-architect - 当主题涉及任务中心、交接、检查点、CLI流程或会话生命周期时,优先选择(Granary工作流负责人)角色。
granary-workflow-lead - 当正确性依赖于类型、符号或编译器信息时,优先选择(语义纯粹主义者)角色。
semantic-purist - 当简单的/diff/docs索引可能优于复杂平台时,优先选择
rg(轻量级反对者)角色。lightweight-contrarian - 当token预算、上下文污染或新会话策略至关重要时,优先选择(上下文优化师)角色。
context-economist - 当工具泛滥是主要风险时,优先选择(工作流保守主义者)角色。
workflow-conservative