consensus
Skill1 starsChanged 41 days agoMITView on GitHub
共识模式——多 agent 辩论强共识诊断。当用户要对一个 bug/问题做多 agent 交叉验证诊断、提到"共识"、"consensus"、"让几个工具一起看看"、"多模型会诊"时使用。opencode 作为编排者，通过 ACP 协议调度 dsh/omp/自 spawn 的子 agent（不同模型）盲辩→交叉批评→表决，产出全员一致的根因共识+分歧报告，支持分级追问。
3 other projects use a version of this.
Adopted by
Most recently taken up by
- spscream/ai-skillstheir own version, first seen Oct 13, 2026
- spsk-dev/consensustheir own version, first seen Mar 31, 2026
- ibahgat/oh-my-iflowtheir own version, first seen Mar 14, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- ibahgat/oh-my-iflow★ 2same name
- spscream/ai-skills★ 1same name
- spsk-dev/consensus★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
Files
Consensus — 共识模式
多 agent 辩论强共识诊断。核心思想：单模型考虑不全，多模型独立盲辩再交叉验证——相同点是共识种子，不同点是辩论议题。
何时触发
用户显式要求：/consensus <问题> 或明确说"共识模式诊断这个 bug"。
不要自动触发——三 agent 辩论约 3-12 倍单次调用成本。
快速基线（决策 17）：/consensus --quick <问题>（或明说"快速共识"）——R1 盲辩 + 直接表决，跳过交叉批评轮，约 1/3 成本。适合成本敏感的问题、或与 full 模式做同题对照（回应 Debate-or-Vote 的鞅批评）。
你（opencode 主会话）的角色：协调者，不是辩手
你不辩论、不裁决。你只做：调度辩手、汇总立场、read/grep 实证、起草共识文案、渲染报告。 结论由三个调查员（辩手）表决产生。
执行流程
严格按 SKILL_BODY.md 的六阶段协议执行。关键纪律：
- 盲辩纪律：R1 任务书 = 用户问题原文 + 工作区路径 + 报告 schema 要求，零他方信息
- 槽位 spawn 化：三个辩手全部是独立进程/子 agent，你不得亲自下场
- 证据门槛：保留必须引用具体证据/反例，否则丢弃；反对必须带反驳证据(file:line)并指认分歧点，否则降级为保留
- 增量上下文：R2+ 只发新增材料（他方结论+分歧点），不重贴全文
- 分层判定：根因=全员非反对；修复方案=无反对票，允许多候选并列
- 僵局降级：轮数耗尽仍有实质分歧 → 输出已共识部分+分歧报告（按层标注），交用户裁决，不强制收敛
- 成本披露：报告含调用次数/token/耗时 + 单 agent 对照基线估算（数据源：state.json checkpoint 的 cost 累积）
- 席位预检、永不静默降级：辩论前
preflight；席位运行中失败走三分类诊断（launch/budget/structural），通道回落必披露（channelFallback）；任何收窄调查员独立审查范围的有损动作须用户批准——重试只做无损动作
阵容（默认，可被 .consensus.yaml / ~/.config/consensus/config.yaml 覆盖）
| 槽位 | 宿主 | 模型 |
| 1 | opencode spawn 子进程（opencode run） | null=随 opencode 默认（配置可钉死） |
| 2 | dsh acp profile | 随 dsh profile（配置标注用） |
| 3 | omp acp | null=随 omp 默认 |
席位二进制经三级解析：CONSENSUS_BIN_{OPENCODE,DSH,OMP} env > 配置 roster[].bin > PATH 探测（lib/preflight.mjs）。异构性纪律：三席刻意选不同模型，避免同模型趋同——设计依据见 README「协议设计依据」。
追问分级
- 默认：基于共识记录直接回答（零调查员调用）
- 追问引入新信息或挑战已有共识 → 向相关调查员（ACP 会话续聊）发起一轮定向辩论，重过阶段 3-5 增量流程
工具与文件
- 通道客户端：
lib/acp-client.mjs（ACP 协议）+lib/cli-client.mjs（CLI 回落通道） - 调查员定义：
lib/investigators.mjs（阵容组装、配置解析、凭据读取） - 工程纪律：
lib/preflight.mjs（席位预检/二进制三级解析）+lib/diagnose.mjs（失败三分类）+lib/checkpoint.mjs（逐轮 checkpoint） - 配置：
~/.config/consensus/config.yaml全局默认 + 仓库.consensus.yaml覆盖（样例见config.example.yaml） - 落盘：
.consensus/runs/<timestamp>/（state.json=checkpoint 与阵容快照 + reports/ + report.json + report.html + events.ndjson） - 完整协议细节：
SKILL_BODY.md
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.