Claudeforce: 37 Sales Skills for GTM Engineers

On August 26, 2026, Salesforce and Anthropic announced Claudeforce, releasing the Salesforce in Claude open beta on September 15, 2026. At Drevon, we monitor how foundational AI models integrate with enterprise revenue infrastructure. For revenue teams building outbound workflows with our free Mac desktop app, this partnership establishes standard CRM orchestration inside conversational interfaces while reinforcing the line between internal database management and live external account research.

During initial rollouts, approximately 7,000 internal Salesforce sellers deployed the integration alongside enterprise pilot organizations including GitLab, Siemens, Deloitte, and Legora. This guide examines the technical architecture behind Claudeforce, audits its 37 prebuilt sales skills, breaks down the Flex Credits consumption economics, and outlines how GTM engineers connect internal CRM execution with live edge intelligence.

---

TL;DR

• Standardized MCP Architecture: Claudeforce replaces bespoke prompt wrappers with 37 bi-directional sales skills operating over Salesforce's AIforce harness via hosted Model Context Protocol (MCP) servers.
• Enterprise Security Boundaries: Claude processes data through Amazon Bedrock within Salesforce's Trust Boundary under zero data retention (ZDR) policies, requiring explicit human authorization before executing CRM write-backs.
• Flex Credits Economics: Execution is billed via Salesforce Flex Credits ($500 per 100,000 credits; 20 credits or $0.10 per standard tool action, 30 credits or $0.15 for voice actions), with 100,000 credits included for Enterprise Edition customers via Salesforce Foundations.
• The Internal Data Wall: Claudeforce evaluates only data stored within Salesforce Data Cloud (Data 360) and CRM objects. It does not browse external websites, monitor live regulatory filings, or extract intent signals from developer forums.
• Hybrid Revenue Stack: GTM engineers use local browser agents to gather verifiable external intent signals, streaming structured payloads into Data Cloud via OpenAPI 3.0.x schemas for downstream Claudeforce skills.

---

1. The Claudeforce Architecture: What Shipped and How It Works

Claudeforce is a two-way integration framework split across two primary operational layers: Salesforce in Claude and Claude in Agentforce.

Rather than relying on unstandardized REST API wrappers or custom middleware, Salesforce and Anthropic deployed hosted Model Context Protocol (MCP) servers connected directly to the Salesforce AIforce enterprise harness. According to technical analysis by Apex Hours, this design exposes complex Salesforce endpoints as deterministic, schema-validated tool definitions that Claude invokes natively.

┌─────────────────────────────────────────────────────────────┐
│                       User Interface                        │
│       (Claude Web / Desktop / Cowork / Slack Deal-Rooms)    │
└──────────────────────────────┬──────────────────────────────┘
                               │ MCP Protocol (JSON-RPC)
┌──────────────────────────────▼──────────────────────────────┐
│                  Salesforce AIforce Harness                 │
│   (Object-Level Security, Field-Level Permissions, SSO)    │
└──────────────────────────────┬──────────────────────────────┘
                               │
       ┌───────────────────────┴───────────────────────┐
       ▼                                               ▼
┌─────────────────────────────┐         ┌─────────────────────────────┐
│    Atlas Reasoning Engine   │         │    Salesforce Data Cloud    │
│  (Claude on Amazon Bedrock) │◄───────►│  (DSOs → DLOs → DMO Models) │
└─────────────────────────────┘         └─────────────────────────────┘

The system operates across three core architectural components:

1. Salesforce in Claude Plugin: An interface available across paid Claude plans (Pro, Team, Enterprise) that exposes 37 prebuilt sales skills directly in conversational workspaces.
2. Claude in Agentforce: Anthropic's Claude serves as a reasoning engine for the Atlas Reasoning Engine, orchestrating autonomous planning inside Agentforce Vibes and Agentforce Coworker across Salesforce and Slack.
3. Enterprise Governance and Permissions: As detailed by TechNode Global, the integration inherits existing Salesforce user permissions, Profiles, Field-Level Security (FLS), and sharing rules rather than requiring separate API tokens.

As reported by MarTech, this decoupled architecture allows revenue teams to interact with CRM data through a conversational interface while preserving standard Salesforce data governance.

---

2. The 37 Prebuilt Sales Skills: Workflow Orchestration

The 37 prebuilt sales skills in Claudeforce standardize multi-step operational routines across six revenue domains. Unlike low-level platform primitives (such as invocable Apex methods, Flow actions, or standard updateRecord CRUD endpoints), these skills represent high-level, workflow-oriented capabilities co-engineered for conversational execution.

As documented by Unite.AI and cataloged by AI Weekly, the functional taxonomy spans key sales cycles:

Workflow Domain	Prebuilt Skills Included	Primary CRM Objects Touched	Core Operational Output
Pipeline & Forecasting	Daily Briefing, Pipeline Review, Deal Signals, Team Rollup, Forecast Narrative, Deal-Slip Scenarios, Weekly Recap	Opportunity, Pipeline, ForecastItem	Structured pipeline health metrics, risk summaries, and quota coverage analysis.
Deal Strategy & Execution	Deal Review, Stakeholder Mapping, Close Plan Builder, Objection Handling, Deal-Advance Gap Analysis, Opportunity Field Updates	Opportunity, ContactRole, Task	Mutual action plans, champion alignment maps, and stage progression write-backs.
Prospecting & Inbound Growth	Prospect Research, Lead Triage, Account Tiering, Lead Routing, Prospect Outreach Drafting	Lead, Account, CampaignMember	ICP match scores, inbound lead assignment, and context-grounded outreach drafts.
Meetings & Communications	Call & Meeting Prep, Call Follow-Up, Transcript Summarization, Calendar Context, Slack Deal-Room Sync	Event, Task, VoiceCall, ContentDocument	Pre-call briefing dossiers, action item extraction, and Slack channel updates.
Renewals & Customer Success	Renewal Prep, Customer Health Review, Customer Feedback Aggregator, QBR Deck Builder	Contract, Asset, Case, UsageData	Churn risk indicators, account health summaries, and QBR slide outlines.
CRM Hygiene & Personalization	Win/Loss Analysis, Activity Logging, CRM Hygiene Auditing, Automated Rep Setup	OpportunityHistory, ActivityMetric, User	Stale opportunity flags, missing contact role audits, and territory onboarding artifacts.

These workflow skills eliminate the need for RevOps teams to build custom middleware for routine internal queries. However, write operations remain governed by human confirmation gates to prevent unverified data from altering CRM stages autonomously.

---

3. The Commercial Framework: Flex Credits Breakdown

Deploying Agentforce and Claudeforce workflows requires an understanding of Salesforce's consumption model. Salesforce bills agent execution through Flex Credits, which act as a unified currency across Agentforce actions and Data 360 operations.

As outlined in pricing documentation analyzed by Enterprise Dreamin', Salesforce structures autonomous execution costs around discrete platform actions:

                          ┌───────────────────────────┐
                          │   Agentforce Billing Path │
                          └─────────────┬─────────────┘
                                        │
           ┌────────────────────────────┴────────────────────────────┐
           ▼                                                         ▼
┌──────────────────────────────────────┐  ┌──────────────────────────────────────┐
│      Flex Credits (Consumption)      │  │        Conversations SKU (Flat)      │
├──────────────────────────────────────┤  ├──────────────────────────────────────┤
│ • $500 per 100,000 credits           │  │ • $2.00 per 24-hr conversation       │
│ • Standard Action: 20 credits ($0.10)│  │ • Best for multi-turn service bots   │
│ • Voice Action: 30 credits ($0.15)   │  │ • Single billing meter per org       │
│ • Break-even: 20 actions/session     │  │ • Predictable high-volume customer UI│
└──────────────────────────────────────┘  └──────────────────────────────────────┘

The financial analysis from NPI Financial breaks down the underlying unit economics:

1. Flex Credits Baseline: Organizations purchase Flex Credits in packs of $500 per 100,000 credits ($0.005 per credit).
2. Action Unit Costs: A standard backend action (such as executing a SOQL query, updating a record, or running a Flow) consumes 20 credits ($0.10). Voice-based agent actions consume 30 credits ($0.15), and custom LLM prompts consume 2 credits ($0.01). Passive conversational turns without tool calls consume 0 credits.
3. Break-Even Mechanics: The flat $2.00 per conversation SKU breaks even with Flex Credits at 20 standard actions per session ($0.10 × 20 = $2.00). Internal sales workflows that require fewer than 20 discrete tool executions per session operate more economically under the Flex Credits meter.
4. Foundations Allowance: Organizations on Salesforce Enterprise Edition and above receive an initial allotment of 100,000 Flex Credits at $0 through Salesforce Foundations to build and test Agentforce workflows.

---

4. Why CRM-Native Agents Hit the "Internal Data Wall"

Claudeforce reasons over internal records, but its operational scope is strictly bounded by the data present in Salesforce Data Cloud.

In B2B sales development, critical buying signals occur outside internal databases. Internal CRM records decay over time as prospects change roles, adopt new software, or restructure teams. When Claudeforce executes skills like Prospect Research or Account Tiering, it evaluates historical opportunity records, existing contact objects, or pre-synchronized data tables.

   EXTERNAL REALITY                                INTERNAL CRM BOUNDARY
┌────────────────────────┐                       ┌────────────────────────┐
│ • Live SEC 10-K Filings│                       │                        │
│ • Forum Technical Posts│                       │  Salesforce Data Cloud │
│ • Job Board Tech Stacks│  ──[ The Data Wall ]──►  (Data Lake Objects /  │
│ • Executive Interviews │                       │   Harmonized Models)   │
│ • Live Documentation   │                       │                        │
└────────────────────────┘                       └───────────┬────────────┘
                                                             │
                                                 ┌───────────▼────────────┐
                                                 │      Claudeforce       │
                                                 │  (37 Prebuilt Skills)  │
                                                 └────────────────────────┘

This internal boundary introduces three distinct constraints for GTM engineering:

• Absence of Real-Time Web Exploration: Claudeforce cannot autonomously open a target account's live documentation, pricing page, or engineering blog to verify recent infrastructure migrations.
• Lack of Signal Provenance: Summaries generated purely from internal CRM notes lack external verification URLs, published timestamps, and direct citations, making it difficult for reps to validate claims before outreach.
• Ingestion Latency: Macro changes—such as new executive hires or public technology vendor evaluations—take weeks or months to reach static CRM enrichment tables.

As highlighted by Salesforce Ben, LLM reasoning quality depends directly on context quality. If Data Cloud contains incomplete or outdated account data, the agent generates well-structured proposals based on stale facts.

---

5. The Hybrid GTM Stack: Edge Browser Agents + Claudeforce Execution

To overcome the internal data wall, GTM engineers deploy a split architecture: local desktop browser agents for external discovery, and Claudeforce for internal CRM execution.

In this architecture, edge agents run locally on the engineer's machine. They inspect live corporate websites, regulatory filings, and discussion forums to extract verified buying signals. That structured intelligence is streamed directly into Salesforce Data Cloud via the Ingestion API.

   STEP 1: EDGE DISCOVERY             STEP 2: STREAMING INGEST           STEP 3: CLAUDEFORCE ACTION
┌────────────────────────────┐     ┌────────────────────────────┐     ┌────────────────────────────┐
│   Desktop Browser Agent    │     │   Salesforce Data Cloud    │     │   Agentforce / Claude      │
│                            │     │                            │     │                            │
│ • Browses live career pages│────►│ • Ingestion API (OpenAPI)  │────►│ • Runs Daily Briefing      │
│ • Inspects 10-K filings    │     │ • Maps DSO to DMO model    │     │ • Drafts verified outreach │
│ • Extracts source evidence │     │ • Vector index RAG storage │     │ • Prompts user write-back  │
└────────────────────────────┘     └────────────────────────────┘     └────────────────────────────┘

Ingestion Flow and Schema Specifications

1. OpenAPI Schema Definition: Data Cloud Ingestion API connectors require a flat OpenAPI 3.0.x schema in .yaml format. Schemas must contain no nested arrays or objects, with object names under 80 characters and field names under 40 characters.
2. Streaming Ingestion API Endpoint: Edge agents push structured event payloads via HTTP POST to https://{instance_url}/api/v1/ingest/sources/{connectorName}/{objectName} using OAuth tokens scoped with cdp_ingest_api.
3. Payload Structure: Requests must stay under 200 KB per batch and format records within a top-level "data" array matching the schema fields.

# Data Cloud Ingestion API Schema: ExternalSignals.yaml
openapi: 3.0.1
info:
  title: External Account Signals Ingestion
  version: 1.0.0
paths: {}
components:
  schemas:
    ExternalSignal:
      type: object
      required:
        - signalId
        - accountDomain
        - signalType
        - timestamp
      properties:
        signalId:
          type: string
        accountDomain:
          type: string
        signalType:
          type: string
        sourceUrl:
          type: string
          format: url
        evidenceExtract:
          type: string
        provenanceConfidence:
          type: number
        timestamp:
          type: string
          format: date-time

{
  "data": [
    {
      "signalId": "SIG-2026-90812",
      "accountDomain": "acme-corp.com",
      "signalType": "hiring_expansion",
      "sourceUrl": "https://boards.greenhouse.io/acme/jobs/4829102",
      "evidenceExtract": "Scaling RevOps infrastructure; migrating 500 reps to multi-agent pipelines.",
      "provenanceConfidence": 0.96,
      "timestamp": "2026-09-19T14:32:00Z"
    }
  ]
}

When Claudeforce triggers skills like Prospect Outreach Drafting or Call & Meeting Prep, it pulls this verified stream directly from Data Cloud, generating briefings grounded in live source evidence.

---

6. Decision Rubric: Claudeforce vs. Edge AI Research

Revenue engineering teams should select operational tools based on where the source context lives.

The table below outlines when to route tasks to Claudeforce versus local edge agents:

GTM Task	Recommended Layer	Operational Justification
Inbound Lead Triage	Claudeforce (Lead Triage)	Evaluates internal lead-to-account matches, routing rules, and territory ownership.
Outbound Account Discovery	Edge AI Agents (Drevon)	Navigates live corporate websites, unindexed forums, and interactive job boards.
Daily Forecast Narrative	Claudeforce (Forecast Narrative)	Direct access to historical deal stages, pipeline velocity metrics, and quota coverage.
Competitive Stack Discovery	Edge AI Agents (Drevon)	Identifies recent tooling migrations, active job descriptions, and technical postings.
Pre-Call Meeting Prep	Claudeforce (Call & Meeting Prep)	Synthesizes internal support tickets, past interaction history, and Slack deal threads.
Executive Movement Checks	Edge AI Agents (Drevon)	Scrapes live leadership pages and corporate announcements to verify current titles.
CRM Hygiene & Stage Updates	Claudeforce (Opportunity Field Updates)	Applies deterministic CRM updates that enforce validation rules and field security.

---

Frequently Asked Questions

What is Claudeforce?

Claudeforce is the enterprise integration between Salesforce and Anthropic announced on August 26, 2026. It embeds Claude into Salesforce Agentforce and provides 37 prebuilt sales skills for CRM workflows inside Claude.

What are the 37 prebuilt sales skills in Salesforce in Claude?

The 37 skills are standardized workflow-level capabilities covering pipeline management, deal reviews, prospecting, meeting prep, renewals, and CRM hygiene. They execute via hosted Model Context Protocol (MCP) servers connected to Salesforce's AIforce harness.

How are Agentforce agent actions billed?

Agentforce execution is billed via Flex Credits ($500 per 100,000 credits; 20 credits or $0.10 per standard tool action, 30 credits or $0.15 for voice actions) or flat conversation SKUs ($2.00 per 24-hour session). Organizations on Enterprise Edition receive 100,000 Flex Credits via Salesforce Foundations.

Can Claudeforce replace external web research tools?

No. Claudeforce queries data stored within Salesforce Data Cloud and CRM objects. It cannot independently crawl the live web, verify unindexed job postings, or extract real-time technical buying signals without external ingestion feeds.

How do GTM engineers stream external signals into Claudeforce?

Engineers use local browser agents to extract live web evidence, format the data into flat OpenAPI 3.0.x JSON payloads, and stream records into Salesforce Data Cloud via the Ingestion API for downstream Claudeforce reasoning.

---

Next Steps for GTM Engineers

Claudeforce establishes structured CRM orchestration, reducing the overhead of maintaining bespoke scripts for internal database operations. However, the accuracy of its output depends on the freshness and provenance of the external intelligence fed into Data Cloud.

1. Check Flex Credit Allocations: Confirm your 100,000 Flex Credit allowance in Salesforce Foundations before deploying custom Agentforce actions.
2. Standardize MCP Workflows: Use the 37 prebuilt sales skills for meeting preparation, pipeline reviews, and CRM hygiene checks.
3. Capture Live Buying Signals: Use local browser agents to extract verified web evidence, monitor executive changes, and capture technical buying signals.

Download Drevon for Mac to gather verified intent signals from live web sources. For teams scaling automated account research pipelines across enterprise organizations, explore our enterprise platform.