What are open-source Salesforce skills for AI agents?
A Complete Guide to Kugamon's Public Toolkit on GitHub
Open-Source Salesforce Skills for Claude and AI Agents: A Complete Guide
Salesforce skills for AI agents are packaged instructions — reference guides, scoring rubrics, validation scripts, and workflow steps — that teach a model like Claude how to do real Salesforce work correctly: write production-grade Apex, build Flows that pass review, audit an org, read a debug log, or run a quote-to-cash cycle. Out of the box, an agent connected to your org can call tools; skills give it the judgment to use them the way an experienced admin or developer would.
Kugamon publishes its toolkit publicly at github.com/kugamon. The centerpiece is salesforce-core-skills, an MIT-licensed plugin with 16 general-purpose Salesforce skills, each behaviorally tested against a live org and built to work with any Salesforce MCP server. Around it sit a Salesforce MCP server for Claude Desktop, skills for ISVs and consulting partners, and the Kugamon quote-to-cash skills that teach agents the revenue lifecycle on Kugamon's native objects.
This guide covers what's in the toolkit, what each skill does, how the pieces fit together, when to use them versus Salesforce's own skills library, how to install them in Claude or other agent tools, and why a CPQ vendor publishes this at all.
What's in the Toolkit
| Repository | What it is | License |
| salesforce-core-skills | Claude Desktop / Cowork plugin marketplace with 16 Salesforce admin and developer skills; tool-agnostic (any Salesforce MCP server); also installable in other agent tools via npx | MIT |
| salesforce-mcp-auto-auth-chrome | Local Salesforce MCP server for Claude Desktop with 14 tools; refreshes its session from your Chrome login — no External App, no tokens to paste | MIT |
| salesforce-isv-skills | Reusable skills for Salesforce ISVs and consulting partners | Public |
| kugamon-skills | Plugin that teaches agents the Kugamon object model and the full Quote-to-Cash / Subscription Management lifecycle; builds on salesforce-core-skills | Public, proprietary license |
The 16 Core Skills
Each skill in salesforce-core-skills encodes the rules a senior practitioner applies by habit, several with structured scoring rubrics so reviews come out consistent regardless of who — or what — runs them.
| Skill | What it does | Scoring |
| sf-apex | Create, update, and review Apex classes and triggers — bulkification, governor limits, trigger frameworks | 150-point rubric |
| sf-flow | Create, update, and review Flows — naming, fault paths, entry conditions | 110-point rubric |
| sf-data | SOQL build/optimize/execute, safe bulk DML, test data factories | — |
| sf-lwc | Lightning Web Components development | 165-point SLDS 2 rubric |
| sf-metadata | Metadata creation, org queries, permission set generation | — |
| sf-permissions | Permission set analysis and "who has access to X?" auditing | — |
| sf-diagram | Architecture diagrams (ERDs, OAuth flows, integrations) in Mermaid | — |
| sf-audit | Comprehensive org audit — 18 scored documents in Word, Excel, and HTML from one scan | — |
| sf-test | Generate, review, and run Apex tests; caps assertion-free tests | 120-point rubric |
| sf-security | Security audit plus AppExchange security-review readiness checklist | 100-point rubric |
| sf-debug | Debug log capture and analysis via the Tooling API — trace to root cause | — |
| sf-orgdiff | Org-to-org comparison: drift, release verification, baseline diff | — |
| sf-integration | Named/External Credentials, OAuth flow selection, Platform Events, CDC | — |
| sf-campaigns | Campaign performance, funnels, ROI, lead-source analysis | — |
| sf-leads | Lead/Contact enrichment with citations, confidence ratings, approval gates | — |
| sf-records | Data-health inspection, dedupe/merge plans, safe batched fixes | — |
How the Pieces Fit Together
1. An MCP server gives the agent hands
None of the skill repos install an MCP server; they assume one is connected to your org. Kugamon's salesforce-mcp-auto-auth-chrome is one option — it runs locally for Claude Desktop and refreshes its Salesforce session from your Chrome login, so there's no External App to configure and no token to paste. Any other Salesforce MCP server works too: the skills reason about capabilities (SOQL, DML, metadata, Tooling API), not one vendor's tool names.
2. Core skills give the agent judgment
salesforce-core-skills teaches the agent how a good admin or developer works, and detects one of four execution modes per session — local SFDX metadata, Salesforce CLI, MCP plus code execution, or MCP-only — adapting accordingly. Guardrail hooks stop dangerous writes, and two read-only reviewer subagents (an Apex reviewer and a security auditor) run cheap parallel review passes.
3. Kugamon skills give the agent the revenue lifecycle
kugamon-skills builds on the core plugin for orgs running the Kugamon managed packages. It teaches the object model — opportunities, quotes, orders, order releases, invoices, payments, contracts, subscriptions, renewal opportunities, assets — auto-detects which deployment mode the org runs (CPQ, Quote-to-Cash, Subscription Management, or Subscription Billing), routes New, Expansion, and Renewal record types, populates line items correctly, and knows which amount field to trust. That's why an agent can take "create a quote for Acme Corp for 50 user licenses" or "show me all open renewal opportunities closing this quarter" and get it right on native data — see Why AI Agents Need Native Revenue Data.
Tested, Not Just Written
Prompt files are easy to publish and hard to trust. salesforce-core-skills runs behavioral evaluations against a live Salesforce org — 13 independent agent runs per iteration covering reads, writes with verified cleanup, and analysis, with results published in the repo. The second iteration reported 33 of 39 expectations passed and none failed, with 12 of 13 runs improving over the first. Skills are validated against the current platform (API v66.0, Winter '26 era) with a stated floor of API v60.0, and CI enforces the routing metadata that keeps sixteen overlapping skills from colliding on trigger phrases.
When to Use This — and When to Use Salesforce's Own Skills
Salesforce publishes its own skills library, and the two are complementary rather than competing. The bottom line is about audience: Salesforce's official sf-skills are built for developers — CLI-first, SFDX-project-first, writing new code. Kugamon's skills are built for everyone else — admins, RevOps and sales ops leads, consultants, and ISVs who work in live orgs through an MCP connector and need to audit, fix, secure, monitor, and run the business rather than write code from scratch.
| Use Kugamon's skills when… | Use Salesforce's official sf-skills when… |
| You're an admin, RevOps or sales ops lead, consultant, or ISV — not only a developer | You're a developer writing new Apex, LWC, or Agentforce code |
| You're operating against a live org through an MCP connector — auditing, scoring, debugging, comparing, enriching | You're doing greenfield development inside an SFDX project with CLI-first workflows |
| You need scored, evidence-cited reviews or client-ready audit documents | You need breadth across Agentforce, Data 360, OmniStudio, Commerce, Mobile, and LWR |
| You work in managed-package or subscriber orgs, or you're an ISV preparing for AppExchange security review | You want Salesforce-maintained API currency across the whole platform surface |
| You want org-to-org drift detection, scheduled monitoring, or guardrail hooks | — |
Both together is the intended pattern: developers build in a scratch org with the official library; everyone else audits, tests, secures, monitors, and runs the business in real orgs with this collection.
How to Install
- Claude Desktop / Cowork (recommended): Customize → Marketplace → Add marketplace → enter
kugamon/salesforce-core-skills→ Sync → Install → restart Claude. Repeat withkugamon/kugamon-skillsif your org runs Kugamon. - Any agent tool (Cursor, Codex, OpenCode, and others): the skills follow the open Agent Skills spec —
npx skills add kugamon/salesforce-core-skillsinstalls all 16. - No org yet? The repo ships synthetic sample data so you can try the campaign and lead skills before connecting an MCP server.
Why a CPQ Vendor Publishes This
Kugamon's revenue data lives on native Salesforce objects, and the bet behind the toolkit is simple: the more capable agents get on Salesforce data, the more the native-data advantage shows. Publishing the general skills as open source, and the lifecycle skills publicly, means any customer, partner, or admin can put an agent on their revenue process today — without waiting for a vendor's "coming soon" roadmap, as we discuss in What Is Claudeforce? To our knowledge, no other CPQ vendor publishes agent skills for its own functionality. See Kugamon Quote-to-Cash for the platform those skills drive.
Frequently Asked Questions
Q: What are Salesforce skills for AI agents?
Packaged instructions — reference guides, scoring rubrics, validation scripts, and workflows — that teach an AI agent how to do Salesforce work correctly, from writing bulkified Apex to auditing an org. They sit on top of an MCP server, which provides the tools; the skills provide the judgment.
Q: Are Kugamon's Salesforce skills really open source?
salesforce-core-skills is MIT-licensed. The Salesforce MCP server is MIT-licensed too. kugamon-skills is public on GitHub but under a proprietary license, since it's tied to Kugamon's commercial managed packages.
Q: Do I need Kugamon to use salesforce-core-skills?
No. The 16 core skills are general-purpose Salesforce admin and developer skills that work in any org with any Salesforce MCP server. kugamon-skills is the separate add-on for orgs running the Kugamon packages.
Q: Do the skills work with tools other than Claude?
Yes. They follow the open Agent Skills spec, so tools like Cursor, Codex, and OpenCode can install them with npx skills add kugamon/salesforce-core-skills. The Claude-specific extras — guardrail hooks and marketplace updates — ship through the Claude Desktop install.
Q: How are the skills tested?
Behaviorally, against a live Salesforce org: 13 independent agent runs per iteration covering reads, writes with verified cleanup, and analysis, with eval reports published in the repo. The second iteration passed 33 of 39 expectations with none failed.
Q: How is this different from Salesforce's official sf-skills library?
Different audiences. Salesforce's library is for developers — CLI-first, covering the full platform for greenfield code. Kugamon's skills are for everyone else — admins, RevOps, consultants, and ISVs working in live orgs through MCP: scored reviews, audits, security readiness, drift detection, and revenue operations. The repo recommends using both.
Q: Which Salesforce MCP server should I use?
Any. Kugamon's salesforce-mcp-auto-auth-chrome is a local server for Claude Desktop that refreshes its session from your Chrome login, but the skills reference capabilities generically and work with other Salesforce MCP servers.
Q: Can an agent run quote-to-cash operations with these skills?
With kugamon-skills installed on top of the core plugin, yes — in orgs running the Kugamon packages, an agent can create quotes, convert them to orders, generate invoices, record payments, and surface renewal opportunities, with package detection and record-type routing handled automatically.
Next Steps
Start at github.com/kugamon: add the salesforce-core-skills marketplace, connect a Salesforce MCP server, and ask a real question about your org. If you want the revenue lifecycle on native data as well, explore Kugamon CPQ and Subscription Management, or schedule a demo and bring your agent use case. Contributions to the core skills are welcome — the repo takes PRs.