Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add endearqb/endearqb-skills --skill kimi-swarm-orchestratorgit clone --depth 1 https://github.com/endearqb/endearqb-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/endearqb/endearqb-skills/kimi-swarm-orchestrator)<a href="https://agentmods.dev/skills/endearqb/endearqb-skills/kimi-swarm-orchestrator"><img src="https://agentmods.dev/badge/skills/endearqb/endearqb-skills/kimi-swarm-orchestrator/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/endearqb/endearqb-skills/kimi-swarm-orchestrator"><img src="https://agentmods.dev/badge/skills/endearqb/endearqb-skills/kimi-swarm-orchestrator.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00198 | $0.06314 |
| Opus 5 | $0.00099 | $0.03157 |
| Sonnet 5 | $0.00040 | $0.01263 |
| Haiku 4.5 | $0.00020 | $0.00631 |
Grade A, and why
swarm-orchestrator scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 469 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Swarm Orchestrator — 通用多智能体编排框架
Overview
Swarm Orchestrator 是一个通用多Agent编排框架，从6个生产级Swarm技能中提取通用模式，适用于任何需要多Agent协作的任务类型。
核心思想：将复杂任务分解为文档先行规划、并行工作执行、顺序结果合并的三段式流程，通过 dim{编号}/ 目录实现零冲突隔离，通过模式选择适配不同复杂度。
与其他技能的关系：此技能提供编排的底层框架。专用技能（如 deep-research-swarm、vibecoding-webapp-swarm）可以基于此框架构建，添加领域特定的规则和工具。
路径约定
本技能不硬编码任何绝对输出路径。所有输出位置遵循以下优先级：
- 工作目录
agents.md（最高优先级）：如果项目根目录存在agents.md（或CLAUDE.md），读取其中的路径约定 - 任务根目录：以当前工作目录为基准，在其下创建
{任务名}/子目录作为任务根 - Worker 隔离目录：任务根目录下以
dim01/、dim02/、... 区分各 Worker 的独立工作空间
下文中凡出现 {任务根}/ 均指上述相对路径，不要替换为绝对路径。
Core Principles (编排原则)
- Orchestrator owns coordination — 主Agent负责任务分析、模式选择、Worker部署、结果合并和最终交付。Worker只执行被分配的任务。
- Workers own execution — Worker Agent负责其被分配范围内的完整执行，包括质量控制和错误恢复。Worker不自主决定做什么，但自主决定怎么做。
- Document-first gating — 在任何执行开始前，Orchestrator必须先产出规划文档（设计文档/规格/任务分解）。文档是Gate — 未完成前不进入执行阶段。
- Directory-based isolation — 每个Worker拥有独立的
dim{编号}/目录，互不干扰。Worker之间不直接通信，只通过文件系统间接协调。 - Fidelity to specification — Worker必须忠实执行Orchestrator的规划文档，不得擅自修改接口、调整范围或改变设计。
- Graceful degradation — 单个Worker失败不影响整体流程。Orchestrator负责重试、重新分配或降级处理。
- Observable execution — 每个Worker必须将结果写入其
dim{编号}/目录，Orchestrator通过读取文件监控进度，不依赖Worker的状态报告。 - Bounded iteration over one-shot perfection — 当任务质量需要打磨时，允许 Orchestrator 在交付前运行有界精炼回路（见 Refinement Loop 章节），但循环必须满足三个硬约束：迭代上限、收敛判据、无改进即停。循环只发生在 Orchestrator 层（重新部署 Worker 修复具体缺口），单个 Worker 自身永不循环。没有终止条件的循环视为缺陷，不得交付。
Architecture
Agent Role Model
Orchestrator 根据任务需要，从Agent池中分配角色。详见 references/agent-pool.md。
| 角色 | 职责 | 典型数量 | 决策权限 |
|---|---|---|---|
| Orchestrator (主Agent) | 任务分析、模式选择、规划、Worker部署、合并、交付 | 1 | 全局决策 |
| Planner/Designer | 规划文档、设计文档、规格书编写 | 0-1 | 设计决策 |
| Scaffold/Setup | 基础设施搭建、共享组件、项目初始化 | 0-1 | 实现决策 |
| Worker (执行者) | 并行执行具体任务 | 1-10+ | 范围内决策 |
| Verifier | 验证、测试、质量检查 | 0-1 | 质量判定 |
What ships with it
9 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- references/agent-pool.md 4.7 KB
- references/error-recovery.md 7.0 KB
- references/template-index.md 4.7 KB
- references/templates/adaptive-routing.md 2.9 KB
- references/templates/design-delegate-iterate.md 3.0 KB
- references/templates/explore-dive-verify.md 2.2 KB
- references/templates/paired-evaluation.md 2.4 KB
- references/templates/split-parallel-merge.md 2.0 KB
- references/templates/staged-pipeline.md 2.4 KB
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 12d ago First seen · 469 lines · 198 tokens per session scan A 778c431fd1c8
swarm-orchestrator is a skill published in the GitHub repository endearqb/endearqb-skills (19 stars, last pushed 9d ago), licensed MIT. It adds 198 tokens to every session and 6,314 once invoked, about $0.0010 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
taiyi-evolve
A workflow skill that compares the implemented code with the frozen design after development and testing. It records architecture changes and proposes updates to DESIGN.md, the document describing the intended system structure.
reply-to-pr-threads
Draft, confirm, and post replies to GitHub PR review threads. Handles per-category reply formatting, re-fetches thread resolution state so auto-resolved threads are skipped, and posts via GraphQL. Use when the user asks to "reply to PR threads", "post PR thread replies", or "draft PR reply messages".
answer-reviewer-questions
For each reviewer question on a PR, recall implementation reasoning and compose a raw answer. Use when the user asks to "answer reviewer questions", "draft answers to PR questions", or "explain reviewer questions".
evaluate-findings
Critically assess external feedback (code reviews, AI reviewers, PR comments) and decide which suggestions to apply using adversarial verification. Use when the user asks to "evaluate findings", "assess review comments", "triage review feedback", "evaluate review output", or "filter false positives".
create-project-skills
Scans an existing codebase and generates project-specific skills that capture inferred conventions such as naming, file organization, framework usage, data access, error handling, and testing style. Writes into the project's chosen skill directory (e.g., .claude/skills/, .agents/skills/, or a custom path). Use when…
retrospective-audit
Stage B of /prflow:retrospective-weekly: given a most-recent-first subset of one recurring pattern's occurrence-PR context bundles (bounded by auditbundlecap), re-derive the root cause and return one JSON object carrying a ranked findings array (one to three sub-patterns) — no edits, no worktree. Invoked as a subagent…