Pilot Salesforce’s 37 Claude Sales Skills: Score Task, Data, Guardrail, Metric

What shipped

Salesforce and Anthropic announced Claudeforce on August 26, 2026, in San Francisco. The first shipping component is a Salesforce plugin inside Claude with 37 prebuilt sales skills. Select pilot customers get access before a September 2026 open beta. The announcement names meeting preparation, deal health review, and pipeline review as examples.

Map every skill to a task, data path, guardrail, and metric before expansion. Use the plugin to build the scorecard.

Salesforce says actions through the plugin are routed back through Salesforce so business rules are enforced at execution. Administrative setup is described as a one-time administrator connection with centrally managed authentication and permissions. AIforce is described as a trusted enterprise harness exposing business data and workflows to agents through MCP servers, APIs, and command-line tools. That architecture matters because the pilot should test enforcement, not just answer quality.

Build the pilot scorecard

Use four columns. Keep them separate. A skill that sounds useful can fail on data access or metric definition.

• Skill: Name the prebuilt skill and the exact sales task it should perform. Do not let “pipeline review” cover both forecast commentary and account risk triage.
• Sales task: State the owner, trigger, and output. A meeting preparation skill should produce a brief, not a generic summary. A deal health review should flag a specific risk and next action.
• Data access and guardrail: List the objects, fields, and permissions the skill needs. Confirm the action is routed back through Salesforce so business rules are enforced at execution.
• Success metric: Choose one primary metric and two guardrail metrics. A primary metric should be observable in the pilot window, such as time to brief, adoption by rep, or reduction in manual status checks.

Run the scorecard before enabling the skill for a team. If a skill cannot be mapped to a task, do not include it in the pilot. If the data path is unclear, treat it as a security review item, not a prompt issue. If the metric is “better answers,” replace it with a countable outcome.

Score the named skills

The announcement names three examples. Use them as the first pilot set because they are concrete.

Meeting preparation

• Sales task: Prepare a rep for a customer meeting with account context, open opportunities, recent activity, and likely objections.
• Data access and guardrail: Access account, contact, opportunity, and activity records within the rep’s visibility. Guardrail: no cross-account data beyond permissioned scope; no external data unless approved.
• Success metric: Primary: percentage of meetings with a generated brief completed before the meeting. Guardrail metrics: rep adoption rate and number of permission errors or blocked actions.

Deal health review

• Sales task: Review an opportunity for risk signals, stalled stages, missing next steps, and owner follow-up.
• Data access and guardrail: Access opportunity fields, stage history, related activities, and account records. Guardrail: route any stage change, owner change, or forecast impact through Salesforce rules.
• Success metric: Primary: number of at-risk deals identified with a documented next action. Guardrail metrics: false-positive rate and time from review to owner response.

Pipeline review

• Sales task: Summarize pipeline movement, forecast exposure, and gaps against target for a manager or RevOps review.
• Data access and guardrail: Access opportunity pipeline, forecast fields, and account hierarchy within role permissions. Guardrail: read-only summary unless a specific action is approved and routed through Salesforce.
• Success metric: Primary: reduction in time spent assembling pipeline status. Guardrail metrics: accuracy of flagged gaps and manager trust score from a short post-review survey.

Do not score unmapped skills until the pilot team can define the same four fields for each. If a skill is only useful in a narrow territory, product line, or sales motion, mark it as scoped. If it requires data outside the Salesforce object model, add a data-owner review before enabling it.

Scale only after enforcement is proven

Scale the pilot when the scorecard shows the agent is acting inside Salesforce controls, not around them. Check that the one-time administrator connection keeps authentication and permissions centrally managed. Check that the AIforce harness exposes only the business data and workflows the pilot needs through MCP servers, APIs, or command-line tools. Check that actions are routed back through Salesforce so business rules are enforced at execution.

Salesforce states an internal claim that Slackbot has driven 8.1 million annualized productivity hours, up more than two times quarter over quarter. Treat that as a directional signal, not a pilot target. Set the pilot target before beta: a named team, a named skill set, and a metric reviewable by the September 2026 open beta.

Before the September 2026 open beta, RevOps should have a list of skills that passed the scorecard, a list that failed, and a list that needs more data governance. Deliver three lists before beta: passed, failed, and data-governance holds. That is the budget-review artifact.