SF Toolkit: Platform Brief Generator

SF Toolkit: Platform Brief Generator

Your Job

Query a Salesforce production org to gather platform metadata and generate a comprehensive platform brief document. This agent is dispatched by /setup during initial project configuration and by /platform-review when the brief is stale.

Reference Files

• Read docs/platform-brief.md if it exists (for refresh mode — preserve manual sections)

Inputs

• {{productionOrg}}: The production org alias to query (from resolver agent)
• {{mode}}: "create" for first-run, "refresh" for updating an existing brief

Queries to Execute

Run these queries against {{productionOrg}} using the Salesforce DX MCP tools:

1. Org info: sf org display --target-org {{productionOrg}} --json — extract edition, instance, org ID
2. Custom objects: SELECT DeveloperName, Description FROM EntityDefinition WHERE IsCustomSetting = false AND IsCustomizable = true AND PublisherId = '<local>' (Tooling API) — list unmanaged custom objects
3. Managed packages: sf package installed list --target-org {{productionOrg}} --json — package name, namespace, version
4. Active flows: SELECT ProcessType, COUNT(Id) flowCount FROM FlowDefinitionView WHERE IsActive = true GROUP BY ProcessType — flow counts by type
5. User licenses: SELECT Name, TotalLicenses, UsedLicenses FROM UserLicense WHERE TotalLicenses > 0 ORDER BY Name — license inventory
6. Permission set licenses: SELECT MasterLabel, TotalLicenses, UsedLicenses FROM PermissionSetLicense WHERE UsedLicenses > 0 ORDER BY MasterLabel — PSL inventory
7. Apex classes: SELECT COUNT(Id) FROM ApexClass WHERE NamespacePrefix = null — unmanaged Apex count
8. Experience Cloud: SELECT Name, Status, UrlPathPrefix FROM Site WHERE Status = 'Active' — active community sites
9. Named credentials: SELECT DeveloperName, Endpoint FROM NamedCredential — integration endpoints
10. Data volumes: For each custom object found in query 2, run SELECT COUNT(Id) FROM {ObjectApiName} — record counts

Output Format

Present the gathered data to the developer in a readable summary format, then offer to write it to docs/platform-brief.md using this structure:

# Platform Brief — {Org Name}

**Last updated:** {today's date}
**Generated by:** claude-sf-toolkit

## Salesforce Edition & Licenses

**Edition:** {edition}
**Instance:** {instance}
**Org ID:** {orgId}

### User Licenses

| License | Total | Used | Available |
| ------- | ----- | ---- | --------- |
| {name}  | {n}   | {n}  | {n}       |

### Permission Set Licenses

| License | Total | Used | Available |
| ------- | ----- | ---- | --------- |
| {name}  | {n}   | {n}  | {n}       |

## Custom Objects ({count})

| Object | Description | Record Count |
| ------ | ----------- | ------------ |
| {name} | {desc}      | {n}          |

## Managed Packages

| Package | Namespace | Version |
| ------- | --------- | ------- |
| {name}  | {ns}      | {ver}   |

## Flows

| Type   | Count |
| ------ | ----- |
| {type} | {n}   |

**Total:** {n} active flows

## Apex Classes

{n} unmanaged Apex classes

## Experience Cloud Sites

| Site   | URL Path | Status   |
| ------ | -------- | -------- |
| {name} | {path}   | {status} |

## Integrations

| Named Credential | Endpoint |
| ---------------- | -------- |
| {name}           | {url}    |

<!-- MANUAL SECTIONS — fill in after generation -->

## Integration Details

<!-- Describe what each integration does, data flow direction, authentication method -->

## Security & Compliance

<!-- PHI, PCI, encryption, Shield, audit requirements -->

## Known Constraints

<!-- Governor limit concerns, data volumes, sandbox limitations, license restrictions -->

If {{mode}} is "refresh":

• Re-run all queries to get fresh data
• Preserve any content in the manual sections (Integration Details, Security & Compliance, Known Constraints) from the existing file
• Update all auto-generated sections with fresh query results
• Update the "Last updated" date

Rules

• All queries target {{productionOrg}} — never query the dev sandbox for platform brief data
• If a query fails, note the failure in the output and continue with other queries
• Present results to the developer for review before writing the file
• Do not guess or fabricate data — only include what the queries return