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 3338902669-ops/agent-orchestra --skill agent-team-handoffgit clone --depth 1 https://github.com/3338902669-ops/agent-orchestraWrote 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/3338902669-ops/agent-orchestra/agent-team-handoff)<a href="https://agentmods.dev/skills/3338902669-ops/agent-orchestra/agent-team-handoff"><img src="https://agentmods.dev/badge/skills/3338902669-ops/agent-orchestra/agent-team-handoff/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/3338902669-ops/agent-orchestra/agent-team-handoff"><img src="https://agentmods.dev/badge/skills/3338902669-ops/agent-orchestra/agent-team-handoff.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 | $0.00179 | $0.01983 |
| Opus 5 | $0.00089 | $0.00992 |
| Sonnet 5 | $0.00036 | $0.00397 |
| Haiku 4.5 | $0.00018 | $0.00198 |
Grade A, and why
agent-team-handoff 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 — 90 lines — stays where its author put it; the contents beside it link to each section on GitHub.
多 Agent 协作交接（agent-team-handoff）
把一群 AI 助手变成一支有纪律的团队：谁在干什么、干到哪了、下一步是什么，全部落在文件里，任何成员随时接续。
一、共享目录约定（团队唯一事实源）
在项目外建一个共享目录（示例名 .team-shared/），固定放四份文件：
| 文件 | 内容 | 更新时机 |
|---|---|---|
| CURRENT-TASK.md | 当前任务交接单（状态/负责人/进度/下一步/证据） | 每次实质进展 |
| HANDOFF-RULES.md | 团队协作规则全文（本技能的本地化版本） | 规则变更时 |
| MEMORY-SNAPSHOT.md | 共享记忆快照（决策、偏好、踩坑） | 定期/重大决策 |
| AGENT-ROLES.md | 成员角色表与能力分工 | 编制变化时 |
规则：所有成员只认这份目录，聊天记录不算数。
二、铁律五条
- 先读再动：开始任何任务前，先读 CURRENT-TASK.md。
- 接续不重做：交接单状态为「进行中」时，从记录的当前步骤继续，禁止从头重来。
- 进展留痕：有实质进展立即回写交接单（进度 + 当前步骤 + 下一步 + 证据）。
- 切换先交接：暂停或移交前，把交接写完整，状态改为「等待接续」。
- 完成附证据：改「已完成」必须带验收证据（命令输出 / 截图 / 测试结果），没有证据不许说做完了。
三、任务状态机
待办(queued) → 进行中(in-progress) → 等待接续(handoff-wait) → 进行中 … → 已完成(done)
↘ 受阻(blocked)：写明具体阻塞条件，连续 3 轮未解决才允许标记
四、单写入者边界
- 同一时刻，一个文件只允许一个成员写入；其他人先读后排队。
- 改公共配置（构建脚本、部署配置、依赖清单）前必须先读当前值并声明改动，防止互相覆盖。
- 冲突处理：以交接单里最后一条带时间戳的记录为准。
五、角色分工模板（按能力分派，不是按头衔）
| 角色 | 职责 | 回退条件 |
|---|---|---|
| 总控 | 需求拆解、派发、验收、记忆与交接管理 | — |
| 调度员 | 任务路由、负载均衡、状态监控、超时熔断（优先用脚本/看板等确定性手段，不用 LLM 临场发挥） | 调度延迟高或误派发率上升 |
| 主开发 | 写代码、实现、修复、部署 | 连续失败 2 次切备用通道 |
| 审查员 | 只读评审代码/视觉/安全，不给主线改文件 | — |
| 专项通道 | 处理棘手故障与能力缺口，不与主开发并发改同一文件 | 问题关闭即退出 |
成本优先路由：日常任务走默认主力模型/订阅 → 失败重试一次 → 再失败切备用通道 → 疑难杂症才动用专项通道。不要因为一次小报错就切换主力。
六、记忆同步
- 每个成员的长期记忆各自维护；共享事实只进 MEMORY-SNAPSHOT.md。
- 提供一个同步脚本（示例）：把各成员当日摘要追加合并进快照，保留最近 N 条。
- 教训类内容（踩坑、纠错）优先级最高，必须进快照，防止同一个坑摔两次。
七、最小闭环示例
- 总控接到需求 → 在 CURRENT-TASK.md 写下目标、验收标准、指派主开发，状态=进行中。
- 主开发读完交接单开工 → 完成一个批次 → 回写进度与证据。
- 完成实现 → 总控派审查员只读评审 → 发现问题回写交接单转回主开发。
- 全部验收通过 → 状态=已完成，证据归档，MEMORY-SNAPSHOT.md 记录关键决策。
八、协同调度防翻车指南（八大坑与工程化解法）
适用场景：多个 AI 助手并行推进时的典型翻车点。每次组建多 Agent 团队前，逐条对照检查。
| # | 坑 | 症状 | 工程化解法 |
|---|---|---|---|
| 1 | 上下文爆炸 | 共享对话越滚越长，早期任务约束和验收标准被挤出窗口 | 交接单只存结论与状态（≤200行）；过程性讨论留在各自本地；每批次开始由总控重发「最小必要上下文」 |
| 2 | 幻觉传染 | 一个成员输出错误信息，经共享记忆被全员当作事实扩散 | 共享记忆分区：写入需总控标记「已验证」才可见；未验证内容放隔离区并标注来源与置信度 |
| 3 | 协作死锁 | A 等 B 的产出、B 等 C、C 又等 A，系统永久挂起 | 资源/产出申请按全局固定顺序排队；每个等待必须带超时（默认15分钟），超时自动升级总控裁决 |
| 4 | 状态不一致 | 两人手里的任务状态对不上，重复做或漏做 | 唯一事实源=交接单状态机；改状态即写文件；口头/会话内状态不作数 |
| 5 | 通信风暴 | 成员互相直聊，消息链路数随人数平方膨胀，延迟指数涨 | 星型拓扑：只允许「成员↔总控」通信，禁止成员互发；确需协作由总控转交 |
| 6 | Agent 蔓延 | 按功能细分拆出一堆角色，边界重叠、没人负全责 | 减法原则：新角色上线前先回答「能否并入现有角色的完整闭环」；每个角色必须端到端负责一段可交付物 |
| 7 | 调度靠 LLM 临场发挥 | 路由派发交给大模型自由决定，同一输入两次派发结果不同 | 调度确定性化：看板扫描+固定规则（按标签/负载/顺序）派发；LLM 只做需要理解力的规划，不做机械分发 |
| 8 | 无验收聚合 | 各自交付后没人拼装校验，缺陷在集成时才爆 | 总控设独立验收门禁：对照交接单验收标准逐条核验+运行证据；不合格整批退回，不带病合并 |
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 · 90 lines · 179 tokens per session scan A 26de4981b4d3
agent-team-handoff is a skill published in the GitHub repository 3338902669-ops/agent-orchestra (103 stars, last pushed 19d ago), licensed MIT. It adds 179 tokens to every session and 1,983 once invoked, about $0.0009 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 2026-08-31.
Other skills, from other repositories
dispatch
Use when a task file exists in .hyperflow/tasks/ and workers need dispatching. Fans out parallel workers under per-batch Reviewers, runs a final integration review, and commits per sub-task. Endpoint of the auto-chain — no auto-deploy. Trigger with /hyperflow:dispatch, "run the plan", "execute the task", "build it"…
tasks
Task decomposition — break features into subtasks, assign workers, track progress.
antigravity-agent-manager
Configure and orchestrate parallel agents using the standalone Antigravity 2.0 Agent Manager and Antigravity IDE.
agent-orchestrator
Meta-skill que orquestra todos os agentes do ecossistema. Scan automatico de skills, match por capacidades, coordenacao de workflows multi-skill e registry management.
hyperflow
Use when applying Hyperflow's orchestration doctrine in Codex, Antigravity, Grok, or another single-agent surface. Auto-invoke for non-trivial engineering work: build, implement, add, refactor, debug, fix, review, audit, plan, scope, design, brainstorm, ship, or deploy. Trigger with /hyperflow:hyperflow, "use…
atmos-migration
This skill helps you migrate a repository to Atmos. It covers native Terraform, Terraform Workspaces, Terramate, Terragrunt, Makefiles, Justfiles, and Taskfiles. It gives minimum-disruption paths, file-layout options, workspace mapping, task-to-command mapping, generatehcl/script decomposition, and the remote-state…