Vercel 推出 Slack Agent Skill 簡化建置

🔑 增強重點摘要

• •Vercel Agent Skills ecosystem partners with Snyk for automated security scanning at submission time, featuring critical-level detectors with 90-100% recall on malicious skills and 0% false positives on legitimate ones[1].
• •Skills directory includes categorized packages like AI SDK skills (e.g., ai-sdk for RAG systems), Deployment skills (e.g., vercel-deploy supporting 40+ frameworks), and Workflow skills for resilient async functions[3].
• •Vercel integrates skills into VS Code extensions and slash commands, enabling agents to access codebase context via agent/skills directories and extension integration points[2].
• •Slack Agent Skill builds on Vercel's internal use of Slack agents for sales automation, such as inbound lead qualification and personalized email drafting[7].

🛠️ 技術深入

• •Skills installer uses npx skills CLI, triggering Snyk's high-throughput API for submission-time scanning across 8 security policy categories, including novel natural language attack vectors like prompt injection[1].
• •Security features include continuous monitoring with re-evaluation as threats evolve, surfaced via 'Security Verified' badges on skills.sh pages[1].
• •Tool Router integration via Composio's MCP handles authentication and execution for 500+ apps (e.g., Gmail, Slack, GitHub) without bloating context windows[4].
• •Intent Layer skill generates hierarchical AGENTS.md files to provide codebase mental maps, reducing Claude's token usage from 40k to 16k and improving bug detection[4].

🔮 前景展望AI analysis grounded in cited sources

⏳ 時間線