If you're building or debugging Agentforce agents with Agent Script, this skill keeps Claude grounded in the actual syntax and toolchain. It covers the full lifecycle: writing .agent files, generating backing logic, using the sf CLI with the right flags (always add --json), and reading session traces to diagnose why your agent routed to the wrong subagent or failed to invoke an action. The standout rule is "diagnose before you fix," which forces Claude to preview with live actions and read trace files before touching code. Agent Script shipped in 2025 with zero training data in any model, so without this skill Claude will hallucinate syntax from JavaScript or Python. Worth using if you're doing anything beyond basic Apex or Flow work.
npx skills add https://github.com/forcedotcom/sf-skills --skill developing-agentforce