← 返回 Skills 市场
csuwl

acpx-team

作者 Wang Lei · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ✓ 安全检测通过
149
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install acpx-team
功能描述
Multi-agent collaboration and task delegation via the Agent Client Protocol (ACP) using acpx. Form agent teams from Claude Code, Codex, OpenCode, Gemini, Cur...
使用说明 (SKILL.md)

acpx — Multi-Agent Collaboration via ACP

Form teams of AI coding agents, run deliberations, build consensus, and delegate work — all through the Agent Client Protocol.

Prerequisites

npm i -g acpx@latest

Install the target agents you want to use (e.g., npm i -g @anthropic-ai/claude-code).

Quick Start: Single Agent

# One-shot (no session state)
acpx claude exec "fix the failing tests"

# Persistent multi-turn session
acpx claude sessions new --name worker
acpx claude -s worker "analyze the auth module"
acpx claude -s worker "now refactor it based on your analysis"

Quick Start: Multi-Agent Council

The fastest way to get multiple opinions on a task:

# 1. Assemble team (create named sessions for each agent)
acpx claude sessions new --name claude-r && acpx codex sessions new --name codex-r && acpx gemini sessions new --name gemini-r

# 2. Round 1: fan-out the same question to all agents in parallel
acpx --format quiet claude -s claude-r "Review src/auth.ts for security vulnerabilities. Be specific." > /tmp/r1-claude.txt &
acpx --format quiet codex -s codex-r "Review src/auth.ts for security vulnerabilities. Be specific." > /tmp/r1-codex.txt &
acpx --format quiet gemini -s gemini-r "Review src/auth.ts for security vulnerabilities. Be specific." > /tmp/r1-gemini.txt &
wait

# 3. Round 2: each agent sees all responses and revises
acpx --format quiet claude -s claude-r "Other reviewers said:\
[Codex]: $(cat /tmp/r1-codex.txt)\
[Gemini]: $(cat /tmp/r1-gemini.txt)\
\
Revise your assessment." > /tmp/r2-claude.txt &
acpx --format quiet codex -s codex-r "Other reviewers said:\
[Claude]: $(cat /tmp/r1-claude.txt)\
[Gemini]: $(cat /tmp/r1-gemini.txt)\
\
Revise your assessment." > /tmp/r2-codex.txt &
acpx --format quiet gemini -s gemini-r "Other reviewers said:\
[Claude]: $(cat /tmp/r1-claude.txt)\
[Codex]: $(cat /tmp/r1-codex.txt)\
\
Revise your assessment." > /tmp/r2-gemini.txt &
wait

# 4. Synthesize (the orchestrator agent does this)
echo "=== Final Reviews ===\
\
[Claude]: $(cat /tmp/r2-claude.txt)\
\
[Codex]: $(cat /tmp/r2-codex.txt)\
\
[Gemini]: $(cat /tmp/r2-gemini.txt)"

Council Protocol

The standard multi-agent collaboration pipeline:

ASSEMBLE → BRIEF → DELIBERATE → CONVERGE → EXECUTE → REVIEW → DELIVER
 组建团队   分发问题   交叉讨论    形成共识    分工执行   交叉审查   交付成果

Step-by-Step

1. ASSEMBLE — Create named sessions for each agent:

acpx claude sessions new --name council-claude
acpx codex sessions new --name council-codex
acpx opencode sessions new --name council-opencode

2. BRIEF — Send each agent the task with a role prefix:

acpx --format quiet claude -s council-claude "[ROLE: Architect] Design a caching layer for our API. Consider: invalidation strategy, TTL, cache-aside vs write-through." > /tmp/r1-claude.txt
acpx --format quiet codex -s council-codex "[ROLE: Performance Expert] Design a caching layer for our API. Consider: invalidation strategy, TTL, cache-aside vs write-through." > /tmp/r1-codex.txt

3. DELIBERATE — Each agent reviews all other responses and revises:

ALL_RESPONSES=$(echo "[Claude]: $(cat /tmp/r1-claude.txt)\
\
[Codex]: $(cat /tmp/r1-codex.txt)")
acpx --format quiet claude -s council-claude "Other architects said:\
$ALL_RESPONSES\
\
Revise your design. Address any concerns raised." > /tmp/r2-claude.txt
acpx --format quiet codex -s council-codex "Other architects said:\
$ALL_RESPONSES\
\
Revise your design. Address any concerns raised." > /tmp/r2-codex.txt

4. CONVERGE — Orchestrator synthesizes a consensus plan.

5. EXECUTE — Delegate implementation to agents in parallel:

acpx --approve-all claude -s exec-claude "Implement the caching layer based on this design: $(cat /tmp/consensus.txt)" &
acpx --approve-all codex -s exec-codex "Write tests for the caching layer: $(cat /tmp/consensus.txt)" &
wait

6. REVIEW — Cross-review implementations:

acpx --format quiet codex -s council-codex "Review Claude's implementation for bugs and edge cases:\
$(cat /tmp/impl-claude.txt)"
acpx --format quiet claude -s council-claude "Review Codex's tests for coverage gaps:\
$(cat /tmp/tests-codex.txt)"

Quick Council Commands

# Assemble a 3-agent team in one line
acpx claude sessions new --name t1 && acpx codex sessions new --name t2 && acpx gemini sessions new --name t3

# Fire-and-forget fan-out (don't wait)
acpx --no-wait --format quiet claude -s t1 "task..." > /tmp/r1-t1.txt

# Collect all Round 1 results
cat /tmp/r1-*.txt

# Cleanup after deliberation
acpx claude sessions close t1 && acpx codex sessions close t2 && acpx gemini sessions close t3

Team Presets

Pre-configured role assignments for common scenarios. See references/roles.md for full definitions.

Preset Agents Best For
code_review maintainer, perf, testing, security, dx PR reviews, quality gates
security_audit security, skeptic, architect, dx, testing Security-sensitive changes
architecture_review architect, perf, skeptic, maintainer, testing Design decisions, tech debt
devil_advocate skeptic, skeptic, architect, maintainer Go/no-go decisions, proposals
balanced neutral × N General tasks, no specialization
build_deploy architect, testing, maintainer Feature implementation

Supported Agents

Agent Command Install
Claude Code acpx claude npm i -g @anthropic-ai/claude-code
Codex acpx codex npm i -g @openai/codex
OpenCode acpx opencode npm i -g opencode-ai
Gemini CLI acpx gemini npm i -g @anthropic-ai/gemini-cli
Cursor acpx cursor Cursor app
GitHub Copilot acpx copilot npm i -g @githubnext/github-copilot-cli
Pi acpx pi github.com/mariozechner/pi
Qwen Code acpx qwen npm i -g @qwen/qwen-code

Agent Mode Switching

Set working modes (Claude Code example; other agents may vary):

Mode Behavior Use When
plan Plan only, no execution Architecture, analysis
default Ask before changes Standard work
acceptEdits Auto-accept edits Trusted refactoring
dontAsk Auto-accept everything Batch tasks
bypassPermissions Skip all checks CI/automation
acpx claude -s worker set-mode plan
acpx claude -s worker set model opus      # or sonnet

Session Management

acpx claude sessions new --name worker    # create named session
acpx claude sessions ensure               # create if missing
acpx claude sessions list                 # list all
acpx claude sessions show                 # inspect current
acpx claude -s worker sessions history    # recent turns
acpx claude -s worker status              # pid, uptime
acpx claude sessions close worker         # soft-close (keeps history)

Output & Permissions

# Output formats
acpx --format quiet claude exec "task"    # final text only
acpx --format json claude exec "task"     # NDJSON stream

# Permissions
acpx --approve-all claude -s w "task"     # auto-approve all
acpx --approve-reads claude -s w "task"   # auto-approve reads
acpx --deny-all claude -s w "task"        # analysis only

# Lifecycle
acpx --cwd ~/repo claude "task"           # set working directory
acpx --timeout 300 claude "task"          # set timeout (seconds)
acpx --no-wait claude -s w "task"         # fire-and-forget
acpx claude -s w cancel                   # cancel in-flight prompt

Bidirectional Communication

Any agent can call any other agent through acpx:

# From OpenCode → Claude Code
acpx claude -s worker "review src/auth.ts"

# From Claude Code → OpenCode
acpx opencode -s helper "analyze test results"

# From any → Codex → Gemini (chain)
acpx codex -s coder "implement X" && acpx gemini -s reviewer "review: $(cat result.txt)"

Reference Files

  • references/roles.md — All 8 role definitions (security, architect, skeptic, perf, testing, maintainer, dx, neutral) with prompt prefixes for Round 1 and Round 2, plus team presets with agent-to-role mappings.
  • references/protocols.md — 7 collaboration patterns (fan-out, deliberation, role council, adversarial debate, sequential pipeline, DOVA hybrid, DCI structured) with acpx command examples, decision matrix, and cost estimates.

Gotchas

  • Mode not settable at creation: Use set-mode after sessions new
  • session/update warnings: Claude Code adapter may emit Invalid params — harmless
  • Dead session recovery: acpx auto-detects and reconnects, replaying mode settings
  • No direct agent-to-agent messaging: All communication goes through the orchestrator
  • --format quiet is essential for piping: Returns only final text, no tool calls or thinking
  • 2 rounds is optimal: Research shows diminishing returns beyond 2 deliberation rounds
  • Session resume preserves context: Use the same -s name across rounds for continuity
安全使用建议
This skill appears to do what it says: orchestrate multiple agent CLIs via the acpx tool. Before installing or running it, consider: 1) The SKILL.md assumes you'll install npm packages (acpx and agent CLIs) — only install packages from sources you trust and inspect their npm package pages and maintainers; global npm installs can execute code during install. 2) The workflows read local source files (e.g., src/*) and write temporary files under /tmp — ensure you are comfortable with those CLIs accessing repository files and that no sensitive secrets or credentials are included in the data you send to external agents. 3) If you plan to run in sensitive or production environments, test inside an isolated environment (container/VM) first. If you need more assurance, request the upstream package repository or checksums for acpx and the agent clients so you can audit them before installing.
能力评估
Purpose & Capability
The name/description (multi-agent orchestration via ACP) matches the SKILL.md: it documents using the acpx CLI to create sessions, fan out prompts, synthesize results, and manage agent roles/presets. No unrelated credentials, binaries, or config paths are requested.
Instruction Scope
Runtime instructions are explicit about running the acpx CLI, creating sessions, reading local files (e.g., src/auth.ts), writing temporary files under /tmp, and combining outputs. Those actions are consistent with code-review and multi-agent orchestration tasks; there are no instructions to exfiltrate secrets or call unexpected remote endpoints.
Install Mechanism
The skill is instruction-only (no install spec). It recommends installing acpx and agent CLIs via 'npm i -g', which is a normal distribution method. Note: global npm installs run arbitrary package code (postinstall scripts) and fetch from the public registry — this is a general operational risk but coherent with the skill's purpose.
Credentials
The skill declares no required environment variables or credentials. The SKILL.md does not request any secret or unrelated environment access beyond what the acpx CLI itself may require at runtime (not documented here).
Persistence & Privilege
The skill does not request always:true, does not alter other skills' configs, and is user-invocable only. Autonomous model invocation is allowed (platform default) but not combined with any unusual privileges.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install acpx-team
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /acpx-team 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.0
Initial release: multi-agent collaboration via ACP with council protocol, 8 role presets, 7 collaboration patterns
元数据
Slug acpx-team
版本 1.0.0
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 1
常见问题

acpx-team 是什么?

Multi-agent collaboration and task delegation via the Agent Client Protocol (ACP) using acpx. Form agent teams from Claude Code, Codex, OpenCode, Gemini, Cur... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 149 次。

如何安装 acpx-team?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install acpx-team」即可一键安装,无需额外配置。

acpx-team 是免费的吗?

是的,acpx-team 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

acpx-team 支持哪些平台?

acpx-team 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 acpx-team?

由 Wang Lei(@csuwl)开发并维护,当前版本 v1.0.0。

💬 留言讨论