从 AutoResearch 到完整项目的全自动流水线（实验 → 多 Agent 编排 → 技能封装）

This skill is essentially a macro that tells other skills to run and then hands work to an orchestrator agent to 'complete the project.' That is not inherently malicious, but it is powerful and open-ended: the orchestrator could perform network calls, access data, enable or create other skills, or make irreversible changes. Before installing or running it, (1) inspect the implementations/prompts and permissions of the referenced skills (AutoResearch and agency-agents/orchestrator); (2) require manual approvals or breakpoints between stages (do not allow fully autonomous end-to-end runs on your production environment); (3) run it first in a safe sandbox or isolated workspace with no sensitive credentials; (4) restrict what data the orchestrator can access and review audit logs for actions it takes; and (5) avoid providing any secrets or broad credentials unless you have reviewed the downstream agents' behavior. If you can provide the SKILL.md or code for the referenced 'autoresearch' and 'agency-agents' skills, I can give a higher-confidence assessment.

Type: OpenClaw Skill Name: auto-research-pipeline Version: 1.0.1 The skill is a workflow automation tool designed to orchestrate existing research and agent-based skills. It sequentially executes internal commands for setup, execution, analysis, and orchestration (via /autoresearch and agency-agents) as described in SKILL.md. No malicious patterns, data exfiltration, or suspicious instructions were identified.