← 返回 Skills 市场
Openclaw
作者
shineliang
· GitHub ↗
· v1.2.0
· MIT-0
116
总下载
0
收藏
0
当前安装
3
版本数
在 OpenClaw 中安装
/install multi-agent-industry-research
功能描述
Multi-agent collaborative industry research for OpenClaw. Dynamically assigns research roles, runs parallel research via sessions_spawn with codex/gemini/cla...
安全使用建议
Before installing, be aware this skill will: 1) run shell commands that list, move, and may delete directories and files in your working directory; 2) read many RESEARCH_* environment variables at runtime even though none are declared in the registry; 3) call your locally-installed model CLIs (codex/gemini/claude), which will use whatever credentials/config you have for those tools and will see the data you pipe to them; 4) spawn multiple sub-agents automatically and requires certain steps happen in a single turn without pausing. Recommendations: back up and/or run in an isolated workspace, set RESEARCH_CLI_TOOLS=none if you do not want it to call local CLIs, review and (if needed) remove any sensitive files or env vars before running, and inspect the SKILL.md text closely to confirm you accept the described file operations and automatic spawning behavior. If you need stronger assurance, request an explicit declaration of which credentials/CLIs will be used and run the skill first in a sandboxed environment.
功能分析
Type: OpenClaw Skill
Name: multi-agent-industry-research
Version: 1.2.0
The skill bundle implements a complex multi-agent research workflow that utilizes high-risk capabilities, including 'sessions_spawn' for agent recursion and 'exec' for shell-based operations. It specifically uses 'curl' to query a local OpenClaw API (http://localhost:18789/api/session/info) to synchronize model settings and employs shell pipes to send research data to external CLI tools (codex, gemini, claude) for cross-validation as detailed in 'SKILL.md' and 'references/cross-validation.md'. While these behaviors are aligned with the stated purpose of multi-model industry analysis, the extensive use of arbitrary shell execution and local network interaction represents a significant security risk without further sandboxing, although no clear evidence of intentional malice was found.
能力评估
Purpose & Capability
The name/description (multi‑agent research) aligns with use of sessions_spawn, exec, web_search and optional model CLIs. However, the SKILL.md expects the agent to run shell commands that inspect, move, or delete local project directories and to query a local session API — filesystem and local-session access are heavier privileges than a simple 'research report' skill would normally require and should be explicitly justified. Also the skill relies on local model CLIs (codex/gemini/claude) and their credentials implicitly rather than declaring those as required environment variables.
Instruction Scope
The runtime instructions tell the agent to run many shell commands (ls, mv, rm -rf, mkdir, curl to localhost, write/read files under project and /tmp), to create/modify project dirs, and to immediately spawn multiple sub-agents in the same turn (explicitly forbids pausing). It also instructs writing sensitive interim files (/tmp/research-cross-validate.txt) and piping them to third-party CLIs. The skill reads environment variables at runtime (via exec) though none are declared in the registry. These behaviors extend well beyond pure 'analysis' and can modify or delete local data and transmit analysis to external model CLIs.
Install Mechanism
Instruction-only skill (no install spec, no code files). This minimizes supply-chain risk because nothing is downloaded or installed by the skill itself.
Credentials
Registry lists no required env vars, but SKILL.md reads a long list of RESEARCH_* environment variables at runtime (e.g., RESEARCH_MODELS, RESEARCH_CLI_TOOLS, RESEARCH_CLI_TIMEOUT). The skill also expects to use the user's locally-configured model CLIs (which rely on credentials stored in those CLIs) without declaring them. Requesting implicit access to CLI-authenticated provider accounts and reading many env vars is disproportionate unless the user explicitly intends to share those locally configured credentials.
Persistence & Privilege
always:false (normal), but the skill instructs automatic spawning of multiple sub-agents and execution of shell commands (including rm -rf) during a single turn without user confirmation in key steps. Autonomous invocation combined with destructive filesystem operations and piping content to external provider CLIs increases risk. The skill does ask for user confirmation before some destructive actions (archive/delete options), but other parts mandate immediate action and forced sessions_spawn calls.
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-agent-industry-research - 安装完成后，直接呼叫该 Skill 的名称或使用
/multi-agent-industry-research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
v1.2.0
Add history cleanup check (Phase 0 step 1) to prevent old research outputs from polluting new runs; add model inheritance so sub-agents inherit the main session model instead of falling back to openclaw.json defaults; all sessions_spawn calls now require explicit model parameter
v1.1.0
Add OpenClaw metadata (requires, emoji, homepage) for proper installation
v1.0.0
Initial release: multi-agent collaborative industry research with async multi-agent orchestration via OpenClaw sessions_spawn
元数据
常见问题
Openclaw 是什么？
Multi-agent collaborative industry research for OpenClaw. Dynamically assigns research roles, runs parallel research via sessions_spawn with codex/gemini/cla... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 116 次。
如何安装 Openclaw？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install multi-agent-industry-research」即可一键安装，无需额外配置。
Openclaw 是免费的吗？
是的，Openclaw 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Openclaw 支持哪些平台？
Openclaw 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 Openclaw？
由 shineliang（@shineliang）开发并维护，当前版本 v1.2.0。
推荐 Skills