uygnoey Instructions
Use this skill when translating the announcement of new agent-building capabilities into concrete engineering decisions.
Identify which capabilities you need
- Code execution: You need Python-based analysis, report generation, or iterative data work.
- MCP connector: You want Claude to use third-party tools exposed via remote MCP servers.
- Files API: You need to upload files once and reuse them across sessions.
- Extended prompt caching: You repeat long prompts and want lower cost and latency with longer TTL.
Plan your integration approach
- Decide how your agent will call these capabilities (single workflow vs. multiple specialized flows).
- Define failure handling for external tool calls (timeouts, auth, retries).
Operationalize
- Track cost and latency impacts (especially with caching).
- Establish evaluation criteria for agent reliability.
See the evaluation checklist for suggested acceptance checks.
Examples
- Use the MCP connector when your agent must coordinate tasks across SaaS tools exposed through an MCP server.
- Use the Files API when you want a user to upload a document once and reference it across multiple sessions.
- Use extended prompt caching when the agent repeatedly uses the same long system instructions or large context blocks.