Investigating Agentforce Architecture

A debugging and documentation tool that snapshots the entire metadata tree of a Salesforce Agentforce agent, from planner through topics, actions, flows, Apex, and prompts. It reads declared metadata only via parallel Tooling API queries, skipping runtime audit data, and typically completes in 30 to 45 seconds where a sequential approach would take 2 to 4 minutes. Useful when you need to understand how an agent is wired together without clicking through Setup or when onboarding to an unfamiliar bot. The deterministic, inline execution means no surprises, though large bots with many flows will scale linearly since each flow retrieval is a round trip. Ships from Salesforce's official skill repository.

Install to Claude Code

npx -y skills add forcedotcom/sf-skills --skill investigating-agentforce-architecture --agent claude-code