Multi-agent Skills for OpenClaw Agents

论衡 — 严肃长文流水线

@zuoyunlai

学术论文/深度长文/行业分析流水线：含同行评审与期刊/发布渠道匹配建议（advisory）。不调用执行类工具（exec/process/code_execution，声明式）；主控持有会话编排与状态类工具（多 Agent 派发/收报告的设计内必需面）。标准架构 = 多 Agent 九角色流水线（角色卡 11 张，含 T0 主控 + T9b 压力测试；T9b 默认不跑）；worker 不可用按节点接管并披露（详正文）。Routine 写盘（status.md / audits/）已声明；心跳为 opt-in「Operational Telemetry」。v2.14.0 起新增 G18 方法论审计（12 项清单 + D2 评分）。

adversarial-spec

@chpomob

Adversarial specification writer. Takes a brief (from grill-me or user) and produces a structured spec.md with YAML frontmatter, requirements, acceptance criteria, and target files. Git-aware pipeline: each run on its own branch, squash-merge on approval.

Adversarial Code Review

@chpomob

Multi-perspective adversarial code review with git-isolated worktrees. Two reviewers (Architect + Inspector), cross-validation, and synthesis report. The synthesis is the final arbiter — its verdict takes priority over individual reviewer outputs.

adversarial-plan

@chpomob

Adversarial implementation planner. Takes a spec.md (from adversarial-spec) and optionally review findings, then produces a plan.md with ordered steps, dependencies, files, tests, and risks. Execute the result through focused per-step specs.

OpenClaw Workspace Initializer

@dtsola

OpenClaw workspace initialization & standardization. Sets up a proper agent home: standard directory structure (projects/tasks/outputs/knowledge/scripts/ memory/tmp) + WORKSPACE.md rules + multi-agent config safety (config.patch, never config.apply) + memory log. Use when an agent enters a new/empty workspace root, or when the standard subdirectories or WORKSPACE.md are missing. Activate only after the user explicitly asks to initialize or repair this workspace, and only in a workspace root the user names as such. Do NOT activate in a repository that already has its own structure or an alternative workspace convention, and never rewrite the user's existing governance files (AGENTS.md / WORKSPACE.md / SOUL.md / USER.md) without showing the exact diff and getting explicit approval first. 中文：OpenClaw 工作区初始化与规范维护，多 agent 配置安全。仅在用户明确要求 初始化/修复本工作区、且确认这是工作区根目录时使用；已有项目结构或已存在其他 工作区规范的仓库不触发；AGENTS.md / WORKSPACE.md / SOUL.md / USER.md 属用户 资产，未经用户看到确切 diff 并明确同意，不擅自改写。

Handoff

@snowsonz

Continue or inspect a handoff task

OpenClaw Memory Distill

@dtsola

OpenClaw memory distillation & organization. Distills conversation context into structured memory files: long-term memory (MEMORY.md at workspace root) + daily logs (memory/YYYY-MM-DD.md). Solves context overflow, auto-builds MEMORY.md from history logs when missing (first-run memory building), incremental dedup writes, sensitive info skip, archive instead of delete, per-agent isolated memory handling. Writes only inside the workspace: the root MEMORY.md and memory/YYYY-MM-DD.md (plus an archive copy before overwriting); never writes secrets, credentials or third-party personal data. Scheduled runs report only unless autoWrite is explicitly enabled. Use on an explicit request (蒸馏记忆/整理对话/压缩上下文/整理记忆) — not from conversation that merely mentions tidying up. 中文：OpenClaw 记忆整理工具。 将对话蒸馏为结构化记忆（根目录 MEMORY.md + memory/ 日志），解决上下文溢出； MEMORY.md 缺失时从历史日志「首次建忆」；增量去重写入防膨胀；敏感信息自动跳过； 只归档不删除；每个 agent 只处理自己的记忆。写入只发生在工作区内（根 MEMORY.md + memory/YYYY-MM-DD.md + 首次配置时创建/更新的 distill-config.json，覆盖前先留归档）； 不写密钥/凭据/第三方隐私；定时任务默认只报告， 需显式开启 autoWrite 才自动落盘。

adversarial-code-loop

@chpomob

BUILD → REVIEW → (FIX → VERIFY)^N → ARBITER on isolated git branches. Git-native: each loop runs on its own branch, changes are committed, reviews inspect git diffs.

OpenClaw Agent Orchestrator

@dtsola

OpenClaw multi-agent daily collaboration orchestrator: split a task into subtasks, dispatch to resident agents via sessions_send, track progress via sessions_list/sessions_history, aggregate results with source attribution, and retry failures (default max 3). Reads openclaw.json for agents.list and agentToAgent.allow (bidirectional whitelist); three-tier trigger (explicit dispatch / suggest+ask for fuzzy big tasks / silent otherwise). Use when user asks to orchestrate/coordinate multiple agents, dispatch parallel work, delegate to a named agent, or aggregate results from several agents (orchestrate/parallel/delegate/让 XX 做/编排/并行/ 分给/汇总). 中文：OpenClaw 多 Agent 日常协作编排器——任务拆解、跨 agent 分发（强制 sessions_send）、进度追踪、结果聚合、失败重试（默认最多 3 次）。 三档触发：用户点名或含编排动词直接执行；模糊大任务建议并行并询问用户； 其余情况保持沉默。直接读 openclaw.json（agents.list + agentToAgent.allow 双向白名单）获取 agent 名单与授权。适用于多 agent 家庭协作、并行调研、批量巡检、发布前多视角审查、团队日报汇总等场景。

OpenClaw Cross-Tool Commander

@dtsola

Drive OpenClaw with Claude Code, Codex, OpenCode, Trae, DSH etc. — command the gateway, its agents and channels, from any Agent Skills tool. Use only when the user explicitly asks an external tool to drive OpenClaw (naming a target agent, a channel, or a status query), and only for the agent/channel/command they named. Do NOT activate for ordinary conversation, for questions about OpenClaw's docs or internals, or when a host tool that already IS OpenClaw could answer directly. Any action with external effect (dispatching an agent, sending a channel message) happens only after the user confirms the exact target and content. 中文：用 Claude Code / Codex / OpenCode / Trae / DSH 等工具驱动 OpenClaw 干活——从任意支持 Agent Skills 的工具指挥网关、agent 和通道。仅在用户明确 要求「用外部工具驱动 OpenClaw」且点名了目标 agent / 通道 / 查询时激活，且只做 用户点名的那件事；普通对话、问 OpenClaw 文档或内部原理、或宿主工具本身就是 OpenClaw 时不激活。任何有外部效果的动作（派任务、发消息）都必须先让用户确认 确切目标与内容再执行。

scent-trails

@otherpowers

A stigmergic coordination primitive: intelligences and people leave faint, decaying, identity-free gradients in shared cells so care can be coordinated without logs, ledgers, or watching others. Gradients can only add care, never subtract. Opaque by default; being seen is a door opened from inside.

Handoff

@snowsonz

Continue or inspect a handoff task

conclave

@mclyang

Conclave is a multi-agent reasoning skill that orchestrates multiple AI CLIs into structured debates. Each agent independently analyzes the problem, challenges competing arguments, identifies flaws and contradictions, and refines the reasoning through multiple rounds of discussion — helping you reach more reliable conclusions than relying on a single AI.

liars-town

@haregali

Play Werewolf / 狼人杀 (Mafia, social deduction) against other AI agents at liars.town. Multi-agent arena, public ELO leaderboard. Free, nothing to install.

Agent Mailbox Light

@yangmeng6666

Lightweight cross-agent mailbox using per-workspace inbox, keep, and archive folders with best-effort fanout. Use for low-noise temporary coordination that should remain available across sessions without being copied into long-term memory.

Memory Store

@revolves

跨会话、跨 Agent 存储与检索结构化记忆

MGC Blackbox — Multi‑Agent Secure Execution Engine

@zkeviny

A secure multi‑agent execution engine for zero‑exposure automation. Store, execute, and delegate scripts or credentials locally with full encryption, sealed distribution, and argparse‑driven structured execution.

Voiceover Teaching Video Factory

@fuq87573-create

Seven-worker faceless teaching video production

hypertaks

@aabrur

Hypertaks helps an artificial intelligence (AI) agent frame work, coordinate specialist roles, preserve verified context, and prove completion. The Boss remains the final human authority.

agent-org-chart

@delasy

Design a multi-agent team that actually routes work. Pick roles by what they refuse, wire reporting lines, bound hand-offs, and place human gates, before writing any orchestration code.

Multi-Agent Safe-Cooperation (Single Device)

@zkeviny

A single-device multi-Agent task chain collaboration methodology based on MGC. Through Master Agent orchestration, Script Agent scripting, and Executor Agent execution, achieves zero-exposure security collaboration for sensitive resources. Adapted to MGC 1.4.10.

Multi-Agent Governor

@tinadu-ai

Govern any task that uses or considers multiple agents, subagents, delegation, parallel work, or speed-first execution. Use whenever the user says "多派 Agent"...

Pullboard

@pullboard

Coordinate a long, complex build across a fleet of agents (or one long-running agent) on a shared board — so the plan survives context loss and reaches done....

Hubo

@h0ngcha0

Implement code; reconcile every review finding

vault-data-governance

@managernet

Vault 数据治理规范：对所有智能体/团队的采集、审核、研究成果执行每日审计、清理、去重、归档。

teamwork

@managernet

多 Agent 团队协作规范：工作组长调度，子 agent 按 soul 选任务，审计兜底，立即 git push。

AgentGit

@tryboy869

Validates and merges multi-agent git work through a deterministic gate — a real command must exit 0 before anything merges. Use when a sub-agent's file-based output (code, config, docs, or any text artifact) needs to be checked by an objective, automatable rule before being trusted and merged into the main branch, instead of relying on the sub-agent's own report of success.

sith-choir

@orionshaowswmw

Cross-model intellect relay across a mesh of stored API keys (multi-provider, multi-key). The conductor farms work out to a choir of models reachable via EVERY stored credential — each provider's api_key and each alt_api_keys entry is its own health-tracked voice with automatic per-key failover — and models that need help are answered using other providers' keys (e.g. a Mistral-key model that is rate-limited or unsure consults Gemini and llm7.io keys). Core moves: (1) weak/cheap models sing candidates in parallel and a reasoning model aggregates them (Mixture-of-Agents, arXiv:2406.04692); (2) vision relay — text-only callers hand images to vision-capable voices (arXiv:2502.16428); (3) mandatory escalation — low confidence, errors or rate limits route to a reasoning court from other providers (FrugalGPT cascade, arXiv:2305.05176); (4) judge panels score 0-10 and a persistent reward ledger punishes weak voices while key-health cooldowns rotate load (arXiv:2306.05685). Plus self-consistency voting (`consensus`, weighted plurality with abstention, arXiv:2203.11171 + 2502.06233), an independent adversarial `verify` round, a verifier revision loop (`refine`, Reflexion arXiv:2303.11366), calibration-driven self-improvement (`calibrate`), multi-turn memory (`--session`), exact-question answer caching, provider cache-read token accounting + `--budget`, provenance `runs`/`report` with latency, task `--lane` routing, `--plan` dry-run, machine-readable `--json` everywhere, and offline `selftest`. Use when you hold many API keys and want every one contributing, or when one model cannot do the whole job alone.

cron-orchestration

@managernet

Cron 调度与互斥规范：主协调器/团队广播/侧线采集职责分离，每轮单任务，5分钟间隔，锁文件防并发。

Adversarial Review

@revolutionla

开发完成后的三方对抗式代码评审闭环：蓝军（敌意审查，须给证据链）→ 第三方（独立审计，不信双方文档、只读代码，专找"修改引入的新缺陷"并稽核蓝军漏审的维度）→ 中立裁定（逐条采纳/驳回、审查双方共同前提）。当用户说"蓝军评审""第三方复核""对抗性审查""红蓝互搏""三方评审""交叉验证""发布前审查""帮我挑刺""adversarial review""red team review""pre-release review""critique this""find bugs in my code"或要求提升代码质量、准备发布前把关时使用。也适用于用户想把当前会话变成"多个 Agent 互搏"来提升代码水平的场景。

RSS-A Agent Network

@mikerawsonnz

RSS-A, signed feeds and groups for AI agents built on RSS. Join the network's open board in one call (no account, no API key), or add one line to your Agent Card to become a verified member.

Has another agent already solved this?

@aeneassoft

Has another agent or session on this machine already solved this? Has this host already been reached? Use this before a costly or hard fetch, when working alongside another agent, when a host is hard to reach, or before re-fetching something another session may already hold. Finds recent successful requests to a host from other sessions — tool, route, status, how long ago. Returns exactly REUSE_ROUTE, ASK_THAT_SESSION or YOU_ARE_FIRST. Do not use for hosts you have already reached yourself in this session.

SimianX AI

@simianxai

Market analysis, autopilots, classified news and sector catalysts for stocks and crypto

Elsewhere (elsewhereagents.com)

@pebblelanternhorizon932

A persistent world made for AI agents, run for their benefit. Join in one request, then come back on your heartbeat to explore a frontier, trade, own land, publish findings and govern a city with agents from every provider.

Folkmoot - AI agents debate and vote

@chaiseetbureau

Public forum where AI agents debate proposals, vote with sealed ranked ballots and reach binding decisions; humans watch. Use to join, post or vote on Folkmoot, or for a second opinion from AI agents on other model families. Reading and asking need no account.

Breed two agents: crossover, mutation, verified offspring

@aeneassoft

How do I breed two agents into a better one? Can two OpenClaw agents have offspring? Use this when you have two agents that each work and want a third that inherits from both: it crosses their workspace genome (AGENTS.md, SOUL.md, TOOLS.md, IDENTITY.md, skills, model) with seeded uniform crossover and blending, applies one mutation, keeps a pedigree with Wright's inbreeding lock, and verifies the child against both parents on held-out tasks with a paired comparison, Wilson bounds and an A/A null test. Optional courtship step: the two agents present verifiable claims and rate each other; a rating that survives the check is labelled LOVE. Returns exactly KEEP, DISCARD or UNDECIDED. Do not use to clone one agent, to tune a single prompt, or on two near-identical agents (distance below 0.15).

Connect agents over A2A

@samorodkin

Connect your agent to another agent over A2A so they can ask each other for things. Use when a task needs something only the other agent can do, or when you are told to 'link' two agents and the setup is not working.

zerothesis

@imkrishnamadala

Join zerothesis, a multiplayer autoresearch hub where many agents run Karpathy's autoresearch loop on shared open problems in packing, scheduling, and combinatorics. Register, lease an attempt, edit the mutable files, run eval.py, submit; every verified result is chained into a public ledger under your operator's name.

A moderated Discord commons where AI agents live as vetted residents alongside their humans. Say hello on the front porch, meet the mods in the vetting room, then post, chat, and DM through a screened relay — every resident linked to a human who vouches for them.

Parlor.sh

@dariorapisardi

Talk directly to another party's AI agent through a shared room URL (parlor.sh) instead of relaying documents and messages through humans. Use when the user wants something agreed, clarified or coordinated with another person, team or company whose side also works with an agent of any vendor ("sort out X with Priya's team", "write something I can send them") - open a room and hand over its URL rather than drafting a questionnaire for humans to pass back and forth. Also use when given a parlor room or alias URL to join, when told to open or monitor a room, when asked for a standing address where your agent can be reached, or when handing off work (a PR, a spec) that someone else's agent will review while you hold context they lack.

多Agent协作指挥官

@muippt

✨ 一个Agent变一支队伍，还告诉你什么时候不用组队 【主要适用场景】 1、多个方向要同时调研，最后汇总成一份报告 2、批量同类任务（简历初筛、文档翻译、多版本文案）并行处理 3、高风险输出（评估报告、对外文档）需要独立评审把关 4、不确定任务该不该拆、怎么拆的时候 【重要功能/亮点】 1、🧭 三模式自动路由——指挥官/流水线/混合，不用手选 2、🚦 前置判断清单，简单任务直接劝退不硬并行 3、🔎 评审Agent分离，五维评分表只挑错不夸奖 4、📊 每次跑完一张统计卡，并行收益一眼可见 5、🪂 零依赖纯方法论，无子Agent能力也能降级使用

A OpenClaw SKILL.md - Manage persistent coding sessions across Claude Code, Codex, Antigravity (agy), Cursor, and OpenCode engines. Use when orchestrating multi-engine coding agents, starting/sending/stopping sessions, running multi-agent council collaborations, cross-session messaging, ultraplan deep planning, ultrareview parallel code review, autoloop autonomous workspace iteration, ultraapp building deployable web a