← 返回 Skills 市场
multi-role
作者
wzm232803119-arch
· GitHub ↗
· v1.0.1
· MIT-0
165
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install multi-role
功能描述
将 AI 角色拆分为管理与技术层，分别负责协调、需求、编码与测试，实现任务路由、质量关卡和记忆管理。
安全使用建议
This skill appears internally coherent for a governance/multi-role framework and doesn't ask for secrets or external installs. Before enabling it: 1) Confirm the skill's installed path and that you are comfortable with it creating and writing files under that directory (sessions/, task-logs/, .skill-metrics.jsonl). 2) Review OUTPUT-RULES.md — by default the skill is told to be terse and record execution details in logs rather than replies; if you want more upfront transparency, ask the skill to always report file reads/writes in replies. 3) If you or other skills will follow the metrics examples, ensure example paths (like /Users/...) are adapted to the actual installation path so writes do not go to unexpected absolute locations. 4) If you need to limit file I/O, run the skill in a sandboxed environment or review the skill directory permissions first. If you want extra assurance, request a small test run and ask the skill to enumerate the files it will read/write before doing any changes.
功能分析
Type: OpenClaw Skill
Name: multi-role
Version: 1.0.1
The 'multi-role' skill bundle is a highly structured framework designed to orchestrate multiple AI sub-agents (Butler, PM, Dev, QA, CTO) to simulate a professional software development and content creation team. It implements sophisticated 'governance' through role-specific 'constitutions' (e.g., references/constitutions/大管家-角色宪法-SOP.md), standard operating procedures (SOPs), and a three-tier memory management system to optimize token usage and maintain project state. While the framework grants the agent significant file system capabilities (using exec for grep, ls, and sed) and includes instructions to prioritize its internal 'Absolute Prohibitions' over user whims to prevent technical debt (e.g., references/rules/绝对禁令清单.md), these are clearly intended to improve reliability and quality control rather than for malicious purposes. No evidence of data exfiltration, unauthorized remote execution, or harmful prompt injection was found.
能力评估
Purpose & Capability
Name/description (multi-role governance) align with the files and runtime instructions: SOPs, role constitutions, memory routing, and task/log file management. Required env vars/binaries/config paths are none, which is proportionate for a purely governance/instruction skill that operates on files in its own directory.
Instruction Scope
SKILL.md is long and prescriptive: it instructs the agent to create sessions/ and templates/ directories, load constitution and SOP files (via Read tool), write structured task logs and a local metrics file, and follow strict routing/priority rules. These actions are consistent with the stated purpose. Two items to be aware of: (1) OUTPUT-RULES.md is declared highest priority and instructs terseness and to avoid exposing read/details in replies (it also says to answer truthfully if the user asks) — this reduces default transparency about what the skill read or wrote unless the user explicitly asks; (2) SKILL-VALUE-REPORT-SPEC includes examples with absolute user-paths (e.g. /Users/...), so implementations or other skills that copy the examples verbatim might write to those absolute locations if not adjusted to the actual installed path. Both are scope/behavior notes rather than contradictions.
Install Mechanism
Instruction-only skill with no install spec and no external downloads. Low risk from install mechanics because nothing is written to disk by an automated installer; runtime writes are confined to files in the skill repo as described.
Credentials
The skill requests no environment variables, no credentials, and no config paths. Its file operations are described to be limited to the skill directory (sessions/, templates/, references/, archives/). The metrics reporting mechanism is local-file based and optional for other skills to adopt; there are no requests for unrelated secrets or cloud credentials.
Persistence & Privilege
The skill persists data in its own directory (sessions/, task-logs/, archives/) and encourages other skills to append to sessions/.skill-metrics.jsonl. It does not set always:true. The only concern is that the OUTPUT-RULES policy promotes writing execution detail to logs rather than replies (reduces observational visibility) — and the metrics-write examples include absolute paths which could be misused if copied verbatim. Autonomous invocation is allowed (platform default) but not combined with broad credential access here.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install multi-role - 安装完成后，直接呼叫该 Skill 的名称或使用
/multi-role触发 - 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
v1.0.1
Version 1.0.1 of "multi-role-governance"
- No functional or structural changes detected; content remains unchanged from previous version.
- All core mechanisms, processes, and documentation are identical to the prior release.
- This version maintains stability with no additions, removals, or updates to files.
v1.0.0
multi-role-governance 1.0.0
- Introduces a multi-role architecture, separating functions by specialized roles (e.g., 大管家, 产品经理, 研发, 测试, 行政助理) each guided by distinct SOP/宪法文件.
- Establishes strict memory management: project, role, and system-level priorities, with detailed rules for loading, writing, and retaining information.
- Implements a 6-step quality checkpoint system and error-tracking protocol with automated role reconstruction upon repeated errors.
- Adds a parallel task queue managed by 大管家, supporting concurrent user requests with customizable prioritization and dependency handling.
- Codifies mandatory role onboarding, collaboration procedures, and file/directory structures for project sessions, assuring clarity and traceability for every operation.
元数据
常见问题
multi-role 是什么？
将 AI 角色拆分为管理与技术层，分别负责协调、需求、编码与测试，实现任务路由、质量关卡和记忆管理。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 165 次。
如何安装 multi-role？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install multi-role」即可一键安装，无需额外配置。
multi-role 是免费的吗？
是的，multi-role 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
multi-role 支持哪些平台？
multi-role 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 multi-role？
由 wzm232803119-arch（@wzm232803119-arch）开发并维护，当前版本 v1.0.1。
推荐 Skills