gabrielmoreira When to Use This Skill
Use when automating post-install configuration for any Salesforce managed package. This skill reads the package's post-install documentation, discovers available execution methods, and automates the configuration steps — including permission sets, object/field permissions, page layouts, Visualforce page access, and tab settings.
Input
- Required: Package name (e.g.,
LMA,FMA,work.com) - Optional: Path to post-install doc (PDF, markdown, URL)
If no doc is provided, ask the user to supply it.
Workflow
Execute phases in order. Each phase must pass before proceeding.
Phase 1: Discover Available Execution Methods
Priority order:
- Org-native platform MCP servers (highest — direct org access via Headless 360)
- Claude Code external MCP servers (sf-sobject-all, sf-sobject-all-sb, etc.)
- sf CLI fallback (always available if authenticated)
Step 1A: Resolve org API version
Discover the org's current API version dynamically — never hardcode a version number:
sf org display --target-org <alias> --json
From the JSON response, read result.apiVersion (e.g., "67.0"). Store this value and use it as v<apiVersion> in all subsequent REST paths. If the command fails, fall back to the minApiVersion declared in this skill's metadata (67.0).
Step 1B: Check for org-native platform MCP servers
Query the Tooling API for MCP server availability:
sf api request rest "/services/data/v<apiVersion>/tooling/query?q=SELECT+Id,DeveloperName,MasterLabel+FROM+McpServerAccess" --target-org <alias>
Step 1C: Determine execution method
Check which Claude Code MCP tools are available and authenticated.
MCP tool prefixes by org type:
| Org Type | Tool Prefix |
|---|---|
| Production | mcp__sf-sobject-all__ |
| Sandbox | mcp__sf-sobject-all-sb__ |
| Falcon Test (pc-rnd) | mcp__sf-sobject-all-falcon__ |
If MCP needs auth, call the authenticate tool. If auth fails, fall back to sf CLI.
Phase 2: Verify Authentication & Org Identity
- Run a lightweight test query (
SELECT Id, Name, IsSandbox FROM Organization) - If MCP auth fails, automatically fall back to sf CLI
- Display org info and ask user to confirm before proceeding
Phase 3: Verify Package Installation
- Determine the package namespace (ask user if unknown)
- Check via Tooling API (
InstalledSubscriberPackage) — do NOT usePackageLicense - If package not found, stop and inform user
Phase 4: Read and Parse Post-Install Document
Read the provided document and extract discrete configuration steps.
Supported formats: PDF, markdown, URL (via WebFetch), pasted text.
Parsing approach:
- Extract each numbered/bulleted step from the document
- Present the extracted steps to the user for validation before proceeding
Phase 5: Classify Steps & Interactive Plan Review
For each step extracted from the doc, classify as Automated or Manual.
Automation capabilities reference
Via MCP (sobject-all) or sf CLI CRUD:
- Record CRUD on any standard or custom object (PermissionSet, ObjectPermissions, FieldPermissions, SetupEntityAccess, PermissionSetTabSetting, PermissionSetAssignment, etc.)
Via Metadata API retrieve/deploy (sf CLI):
- Page layout modifications (add related lists, fields, sections)
- Profile settings
- Custom metadata type records
Via sf CLI Tooling API:
- Tooling queries (InstalledSubscriberPackage, ApexPage, ApexClass, etc.)
- Any REST-accessible Tooling operation
Manual (no API path — requires Setup UI):
- System permissions not exposed via REST
- Connected app OAuth configuration
- Environment Hub linkage
Interactive approval
Present the classified plan and let the user choose:
- "Approve all" — Execute all steps as planned
- "Let me choose" — Select which steps to approve/skip
- "I have questions" — Discuss specific steps before deciding
Phase 6: Execute Approved Steps
For each approved step, use the resolved execution method.
Execution method reference
| Operation | Via MCP | Via sf CLI |
|---|---|---|
| SOQL query | soqlQuery tool |
sf data query --query "<SOQL>" --target-org <alias> --json |
| Create record | createSobjectRecord tool |
sf data create record --sobject <Object> --values "..." --target-org <alias> --json |
| Update record | updateSobjectRecord tool |
sf data update record --sobject <Object> --record-id <id> --values "..." --target-org <alias> --json |
| Describe object | getObjectSchema tool |
sf api request rest "/services/data/v<apiVersion>/sobjects/<Object>/describe" --target-org <alias> |
| Page layout | N/A | Metadata API retrieve/deploy |
Page layout modifications via Metadata API
Use sf project retrieve start → edit the layout XML → sf project deploy start.
Execution rules
- Idempotency: Before creating any record, query to check if it already exists. Skip if so.
- Report after each step: Show success count, skipped items, and reasons.
- Automatic fallback: If MCP fails mid-execution, retry via sf CLI.
- On failure: Report error, ask user to retry/skip/stop.
Phase 7: Guide Manual Steps (if any)
If any steps could not be automated, present each with Setup navigation instructions. Wait for user confirmation before proceeding to the next.
Phase 8: Summary
Display final summary with step-by-step status, method used, and any skipped items.
Error Handling
- Auth failure mid-execution: Stop, ask user to re-auth, offer to resume
- Duplicate record errors: Treat as "already configured", skip and continue
- Permission errors: Report which permission is missing, suggest resolution
- Unknown step type: Ask user to clarify, offer to mark as manual
Notes
- Priority: org-native MCP > Claude Code MCP > sf CLI > manual
- sf CLI is always a valid fallback for all CRUD and Tooling API operations
- Page layout modifications are automated via Metadata API retrieve/deploy
- Always verify org identity before making changes
- All actions respect the authenticated user's permissions