← ClaudeAtlas

salesforce-agent-scriptlisted

Implement, review, or improve Salesforce Agentforce agents written in Agent Script — the declarative DSL for hybrid reasoning in the Agentforce Builder. Use when working with Agent Script syntax (blocks, expressions, variables, hooks, transitions, action chaining); when designing topics or subagents and the start_agent block; when exposing actions as tools to the LLM via reasoning actions; when balancing deterministic logic instructions against LLM prompt instructions; when migrating between Topics and Subagents terminology (April 2026 rename); when building agents in Agentforce Builder Canvas or Script view; when using Agentforce DX CLI or the Agentforce DX VS Code extension; or when writing agents with Agentforce Vibes. Trigger this skill whenever the user mentions Agentforce, Agent Script, agent topics, subagents, Atlas Reasoning Engine in a scripting context, agent transitions, reasoning actions, @utils, @variables, or @actions — even if they don't explicitly name the skill.
rammc/salesforce-agent-script · ★ 1 · AI & Automation · score 62
Install: claude install-skill rammc/salesforce-agent-script
# Salesforce Agent Script Build, review, and refactor Agentforce agents written in Agent Script. Agent Script is Salesforce's declarative DSL for hybrid reasoning agents on the Agentforce 360 Platform. It compiles to portable JSON metadata consumed by the Atlas Reasoning Engine. This skill covers the language itself and the surrounding pro-code tooling (Agentforce Builder Script view, Agentforce DX CLI, VS Code extension, Agentforce Vibes). It does **not** cover Atlas Reasoning Engine internals, Data Cloud grounding configuration, multi-org deployment strategy, or Apex / Flow / Prompt Template authoring — those are adjacent disciplines. ## Contents - [When to use this skill](#when-to-use-this-skill) - [Mental Model](#mental-model) - [Terminology: Topics vs. Subagents](#terminology-topics-vs-subagents) - [Anatomy of an Agent](#anatomy-of-an-agent) - [The Hybrid Reasoning Decision](#the-hybrid-reasoning-decision) - [Quick Reference](#quick-reference) - [Authoring Workflow](#authoring-workflow) - [Review Checklist](#review-checklist) - [Reference Files](#reference-files) - [Verification Pointers](#verification-pointers) ## When to use this skill Trigger this skill when the task involves: - Writing, reading, or refactoring Agent Script source (`.agent` files or scripts shown in the Builder). - Designing the structure of a new agent: `start_agent` routing, topic decomposition, action surfaces, variable design. - Debugging unexpected agent behavior that comes from prompt