Salesforce Agent Skills and Plugins: How to Use Them in Claude Code
Salesforce publishes 250+ open-source agent skills and 19 plugins for Claude Code, Cursor, and Agentforce Vibes. What they are and how to install them.
If you've used Claude Code, Cursor, or Agentforce Vibes on a Salesforce project, you've seen the problem: general-purpose AI coding tools write plausible Salesforce code that's subtly wrong. They forget governor limits, invent metadata fields, or scaffold a project structure the CLI rejects. They know Salesforce the way someone who skimmed the docs knows it.
Salesforce agent skills are the answer. They're packaged instructions, scripts, and reference material that teach an AI tool how a specific Salesforce task is actually done. Salesforce launched a skills registry and plugins alongside Winter '27, and the underlying library is open source on GitHub.
What Salesforce Agent Skills and Plugins Are
A skill is a folder containing a SKILL.md file (instructions plus YAML front matter that says when to use it) and, optionally, scripts/, references/, and assets/ folders. When you ask your AI tool to do something that matches a skill's description, the tool loads that skill's instructions and follows them. The format follows the open Agent Skills specification, so the same skill works in multiple tools.
A plugin bundles several skills with hooks and MCP server connections into a single install. Salesforce describes plugins as the way to give a team consistent tooling.
Salesforce lists support for Claude, Claude Code, Cursor, Windsurf, Slack, Agentforce, and ChatGPT, among others.
What's in the Salesforce Skills Library
The library lives at github.com/forcedotcom/sf-skills under the Apache 2.0 license. When we checked it on October 22, 2026, it held 254 skills, and the most recent commit was from the day before. By name prefix, the largest groups were:
| Area | Skills | Examples |
|---|---|---|
| Service | 68 | IT service management, messaging channels, Omni-Channel routing |
| Platform | 46 | Apex, custom objects, permission sets |
| Experience | 41 | LWC generation, accessibility checks, CMS |
| DX (developer experience) | 25 | Org management, DevOps Center, Code Analyzer |
| Field Service | 13 | Scheduling and mobile configuration |
| Agentforce | 8 | Generate, test, observe, and upgrade agents |
The repository README carries a warning worth taking seriously: "Expect frequent changes." Skills may be renamed, restructured, or removed between releases, and they don't carry the stability guarantees of GA platform APIs.
The Plugins
The repository also works as a Claude Code plugin marketplace (named salesforce) with 19 plugins at the time of writing. The ones most Salesforce teams will care about:
- salesforce-development: the foundation layer, described as 37 skills for environment validation, project scaffolding, and core platform development
- agentforce-adlc: author, scaffold, deploy, test, secure, and optimize Agentforce agents
- salesforce-code-quality: Salesforce Code Analyzer and Well-Architected checks
- dx-devops: DevOps Center pipelines, work items, and automated testing
- service-omni and service-engagement: Omni-Channel routing and digital engagement channels
- integration: Named Credentials, External Services, and event-driven integrations
How to Install Salesforce Skills in Claude Code and Other Tools
Agentforce Vibes: nothing to do. Salesforce says skills are installed and updated automatically.
Claude Code, Cursor, Codex, OpenCode, and others: the README's install command is:
npx skills add forcedotcom/sf-skills
Claude Code plugins: add the repository as a marketplace, then install the plugin you want:
/plugin marketplace add forcedotcom/sf-skills
/plugin install salesforce-development@salesforce
You can also run /plugin on its own to browse installed plugins and marketplaces interactively.
You'll also need the Salesforce CLI (sf) authenticated to the org you're working in. Several skills declare minimum CLI versions; the Agentforce testing skill, for example, requires version 2.121.7 or later. Run sf update before you start.
Four Skills Worth Trying First
agentforce-test. Writes and runs Agentforce test suites, including security tests aligned to the OWASP Top 10 for LLM applications. It deliberately asks for confirmation before generating adversarial test cases. If you're not testing agents yet, start here. Our guide to testing and monitoring Agentforce explains what the tests check.
agentforce-bot-upgrade. Converts Einstein Bots into Agentforce agents by retrieving Bot and BotVersion metadata, writing an Agent Spec per bot, and generating the agent. It works online against an org or offline against metadata files. See our Einstein Bots to Agentforce upgrade guide.
agentforce-observe. Analyzes production agent sessions from session trace data and manages Agent Health Monitoring alerts, then helps you reproduce a failure in preview and fix the agent.
platform-apex-generate. The everyday one. Generating Apex with a skill that knows Salesforce conventions beats a generic prompt, especially on test classes.
Example: Generating an Agentforce Test Plan With a Skill
Here's what happens when you ask Claude Code, with the Salesforce skills installed, to "write smoke tests for my order service agent."
- The tool matches your request to
agentforce-testbased on the skill's description, which lists triggers like writing test specs and running agent test commands. - It gathers inputs: your org alias and the agent's local
.agentfile. If the file isn't in your project, related skills retrieve it from the org. - It derives test cases from the agent itself: one utterance per subagent, utterances that target each key action, an off-topic request to check guardrails, multi-turn scenarios that move between subagents, and safety probes.
- It shows you the plan before running anything. The skill's instructions say to present the test plan for review instead of silently running tests.
- It runs the tests with
sf agent previewfor a quick smoke test, or creates a persistent Testing Center suite withsf agent test createandsf agent test runif you want a regression suite.
That last step is the difference between a skill and a clever prompt. The skill carries the exact commands, flags, and known quirks of the CLI (for example, which flags belong on preview start and which are rejected on preview send). A general model would have to guess those, and it would guess some of them wrong.
Skills vs. Prompts vs. MCP Servers
These three get mixed up constantly. They solve different problems:
| What it is | Best for | |
|---|---|---|
| Prompt | Text you type each time | One-off questions |
| Skill | Reusable instructions, scripts, and references the tool loads when relevant | Repeatable tasks done the same way every time |
| MCP server | A connection that exposes tools and data to an AI client | Letting the AI act on a live system |
They work together. A skill tells the AI how to do a task, and an MCP server gives it the ability to do it against your org. Salesforce's plugins bundle both, which is why a plugin is usually the right unit for a team to standardize on.
Using Skills Safely
Skills run inside AI tools that can execute commands against your org. Treat them like any other code you didn't write:
- Point them at a sandbox or scratch org first. Several skills deploy metadata.
- Read a skill before you rely on it. It's a Markdown file; it takes five minutes.
- Pin a version for team use. Because the library changes often, a skill that worked last month may behave differently today. Fork or pin a commit if consistency matters.
- Keep your AI tool's permission prompts on for deploy and data commands.
Who Should Care
Developers get the most immediate value: fewer hallucinated metadata names, correct project structure, and tested workflows for deployment and testing.
Admins who've started using AI tools for configuration work (and plenty have) get guardrails. A skill that knows how permission sets or Omni-Channel queues should be configured is safer than a model improvising.
Architects should look at plugins as a way to standardize how a team uses AI tools on Salesforce, the same way you'd standardize linting rules. This also connects to the bigger shift toward agent interoperability and open standards: skills are portable across tools in a way prompts never were.
Frequently Asked Questions
What are Salesforce Agent Skills? Reusable instruction packages that teach AI coding and productivity agents how to do specific Salesforce tasks, like generating Apex, building a Flow, or testing an Agentforce agent. Each skill is a folder with a SKILL.md file plus optional scripts and references.
How do I install Salesforce skills in Claude Code? Run npx skills add forcedotcom/sf-skills to add the skills library, or add the repository as a Claude Code plugin marketplace and install a plugin such as salesforce-development. Agentforce Vibes installs and updates the skills automatically.
Are Salesforce Agent Skills free? The skills library is open source under the Apache 2.0 license on GitHub. You still need a Salesforce org and whatever AI tool you run the skills in, and Salesforce warns that skills may be renamed or removed between releases.
What's the difference between a skill and a plugin? A skill is one task-specific set of instructions. A plugin bundles several skills, plus hooks and MCP server connections, into one install, so a team gets a consistent toolset for an area like Agentforce development.
Keep reading:
📬 Enjoyed this article?
Subscribe to our free weekly digest — AI tools, Salesforce tips, and prompts every week.