Today4 stories · 3 briefs

The Daily Org

Spring '27Sandboxes 6 Feb 2027

Curated by Abhinav (opens in a new tab)

Vol. I · No. 45The independent daily for Salesforce admins, developers and architects4 stories · 3 briefs

Must-readAgentforce and AI

Setting claude's read and write limits in Salesforce through tool restrictions (opens in a new tab)

You will know how the Sales Cloud in Claude plugin reaches Salesforce, why user permissions set what Claude can read, and how the Dispatch tool restriction setting controls whether Claude can create or update records.

Why read this

You will also have a staged rollout plan with a pre-pilot permission review.

  1. Review permissions before the pilot
  2. Set Dispatch to Deny for read-only
  3. Move to Needs Approval for writes
  4. Use Always Allowed only as last step
Salesforce recommends stepping from read-only to approval-gated writes in this order

Sales Cloud in Claude gives sellers a sanctioned way to ask Claude about their Salesforce accounts. The connection runs as the seller, using their own permissions, against live Salesforce data. The Sales Cloud in Claude plugin, built by Salesforce and Anthropic, bundles 37 sales skills, the Salesforce MCP connection and a setup skill that tailors the plugin to each user’s role. The plugin is in beta and requires an enterprise agreement with Anthropic and Sales Cloud licenses.

The plugin works through the standard Salesforce Hosted MCP Server, which exposes Discover, Describe, Dispatch read-only and Dispatch tools. A Salesforce Admin sets up the server and an External Client App, and the Claude admin configures the connector, sets tool restrictions and decides who gets access. When a seller signs in through OAuth, requests run with that user’s credentials, so field-level security and Sharing Rules apply. Reads need no approval, while writes depend on the Dispatch setting.

Because the hosted server has no generic system user, a user’s object permissions, field-level security and Sharing Rules define what Claude can reach. Those permissions were designed for a human, so broad grants such as View All and Modify All may give Claude more than the job needs. Data Claude reads is also sent to the connected model, which makes the review a data-sharing decision as well as a security one.

The Dispatch tool restriction, found under Organization Settings > Connectors, has three options: Needs Approval, which Salesforce recommends as the steady state; Always Allowed; and Deny, which keeps Salesforce read-only. Dispatch can also call deletes directly even though no skill deletes, so Deny or Needs Approval matters. Salesforce suggests a staged approach: clean up permissions before a pilot, agree the Dispatch setting with the Claude admin, then pilot with a small group and review early approvals. Record each decision in an architectural decision record.

Read the original at Salesforce Architects (opens in a new tab)

Worth reading?

RecommendedArchitecture

Salesforce response to a compromised Optimizely script on its websites (opens in a new tab)

You will learn how the Optimizely compromise reached Salesforce web properties, what Salesforce has disabled, and which third-party scripts to audit in your own Experience Cloud, Commerce Cloud and Marketing Cloud pages.

  1. 16 Nov 2026: Exposure window, about 04:00 to 21:00 UTC
  2. 16 Nov 2026: Researcher reports 188 compromised files
  3. 17 Nov 2026: Optimizely confirms the malicious code
  4. Now: Admins audit for Optimizely snippets
The incident ran from the November 16 exposure window through Salesforce's response and the actions it recommends

On November 16, 2026, an attacker exploited a vulnerability in Optimizely, a third-party experimentation platform, and inserted malicious JavaScript into some customer projects. The injected code sat in the Project JavaScript field, which runs on every page. Because the files were still served from Optimizely’s trusted content delivery network, browsers and many security tools treated them as legitimate. Security researcher Scott Helm reported 188 compromised files across 178 organizations, Salesforce among them.

Salesforce says unauthorized changes were made to Optimizely scripts used on certain of its websites, and that those scripts could load malicious code in visitors’ browsers. As a precaution, it disabled Optimizely scripts across preview and production instances, and they will stay off until Optimizely’s remediation is verified, with no timeline to re-enable. Salesforce states it has no evidence the incident affected Salesforce-platform or subscription services. It is following its vulnerability response processes, monitoring for malicious activity, and will notify affected customers if it finds unauthorized access to their data.

The post says the exposure window ran from about 04:00 to 21:00 UTC on November 16. Optimizely says the vulnerability is closed and the injected code has been removed. Its statement points to the account signup flow as the route in.

If you build or host web applications on Salesforce platforms, you are affected only if your own team embedded Optimizely Web Experimentation scripts in them. Salesforce recommends that admins audit Experience Cloud sites, Commerce Cloud storefronts and Marketing Cloud landing pages for Optimizely snippets. Where a customer-facing site used an affected project on November 16, review application logs and watch for payment-form tampering and unusual session activity, such as logins or token use from unfamiliar locations.

Read the original at Salesforce Blog (opens in a new tab)

Worth reading?

DevOps

Building a read-only Apex PMD and security reviewer agent in GitHub Copilot (opens in a new tab)

You will learn how to define a Copilot custom agent that audits Apex for PMD and OWASP-style security issues inside the IDE, without external integrations.

Why read this

You will also see the review output format and the tool restrictions that keep the reviewer read-only.

The article argues that static analysis of Apex usually runs too late, in a CI pipeline that fails a pull request after the developer has finished the work. In locked-down enterprise orgs, adding external AI review bots or webhooks is often blocked by security approval, so the author proposes a review agent that stays inside the existing GitHub Copilot boundary.

The agent is a declarative Markdown file, a .agent.md, stored in .github/agents/ for team sharing or in ~/.copilot/agents/ for personal use. It has a YAML header that gives its name, description and permitted tools, here restricted to search and fetch so the agent cannot modify files. The body is a system prompt that lists the rules to check, such as ApexCRUDViolation, ApexSOQLInjection, OperationWithLimitsInLoop and AvoidHardcodingId, grouped by security, performance, architecture and test quality.

The author gives a sample service class with a concatenated dynamic SOQL string, no sharing keyword, no user-mode enforcement and a DML statement inside a loop. The agent reports each issue with a severity, the rule name, the file and line, a short explanation and a compliant fix. The article closes with three team practices: keep shared conventions in copilot-instructions.md, restrict reviewer tools to read and search, and keep deterministic Salesforce Code Analyzer runs in the pull-request pipeline for audit trails.

Read the original at dev.to (opens in a new tab)

Worth reading?

Agentforce and AI

How Salesforce built ITSM Predictive models before real Customer Incident Data existed (opens in a new tab)

You will learn how the team generated synthetic incident data, validated models using scenario tests instead of accuracy alone, and shipped a reusable Data 360 training and scoring pipeline that trains per customer.

  1. Sync incidents and milestones into DMOs
  2. Daily transform labels milestone violations 0/1
  3. Einstein on Data 360 trains XGBoost classifier
  4. Scheduled batch writes scores to score DMO
  5. AI Accelerator card shows score on incident
How the template turns customer incident history into scored predictions on the incident record

Salesforce’s engineering team set out to predict SLA breaches and major incidents in IT service management without customers needing data scientists, and without any real customer incident history to learn from at the start. They built three Predictive Model Templates: SLA Breach Prediction, Major Incident Prediction, and Change Risk Score. Scores and leading predictors are shown on the incident record through Data 360.

Synthetic data came from product manager samples, expanded by varying features and generated with open-source LLMs within product guardrails. The team noted that the simulated Customer Zero data was insufficient on its own, so grounding in realistic scenarios mattered more than volume.

Because SLA breaches are rare, aggregate accuracy can mislead; a naive model could look 95% accurate and still be useless. Instead, the team held all incident features fixed and varied one factor at a time, such as owners with a higher or lower historical milestone miss rate, checking that the score moved in the expected direction. They also checked whether the top three predictors matched what an experienced service manager would flag. The article stresses that these results validate tested scenarios, not production accuracy.

The template ships the training contract rather than fixed weights: DMO API names, transformation logic and model structure. Each customer’s history determines the learned parameters. Incidents, milestones and related records sync into DMOs, a daily transform labels milestone violations as a 0/1 target, and Einstein on Data 360 trains an XGBoost binary classifier. A scoring transform writes predictions to a score DMO, displayed on the AI Accelerator card.

Two constraints remain. Scoring runs as a scheduled batch about once an hour rather than in real time, chosen for scalability with large customers. Training requires roughly 400 completed incidents, and prediction quality depends on the customer’s history. The templates reached general availability in February.

Read the original at Salesforce Engineering (opens in a new tab)

Worth reading?

In brief