description: "Use this skill when users need to create, generate, modify, or validate Salesforce Lightning pages (FlexiPages). Trigger when users mention RecordPage, AppPage, HomePage, Lightning pages, page layouts, adding components to pages, or page customization. Also use when users say things like 'create a Lightning page', 'add a component to a page', 'customize the record page', 'generate a FlexiPage', or when they're working with FlexiPage XML files and need help with components, regions, or deployment errors. Always use this skill for any FlexiPage-related work, even if they just mention 'page' in the context of Salesforce. DO NOT TRIGGER when users ask about Visualforce pages, Aura components without FlexiPage context, page layout assignments in the UI, or Lightning Web Component development that does not involve placing components on a FlexiPage."
- Understand FlexiPage structure and component configuration
- Work with page layouts or Lightning page customization
- Edit or update ANY *.flexipage-meta.xml file
## Specification
## Overview
**CRITICAL: When creating NEW FlexiPages, you MUST ALWAYS start with the CLI template command.** Never create FlexiPage XML from scratch - the CLI provides valid structure, proper regions, and correct component configuration that prevents deployment errors.
Generate Lightning pages (RecordPage, AppPage, HomePage) using CLI bootstrapping for component discovery and configuration.
---
## Quick Start Workflow
### Step 1: Bootstrap with CLI
**MANDATORY FOR NEW PAGES: This step is NOT optional.** Always use the CLI template command when creating a new FlexiPage. The CLI generates valid XML structure, proper regions, and correct metadata that prevents common deployment errors. Only skip this step if you're editing an existing FlexiPage file.
**`<packageDirectory>`** = the `path` value from `sfdx-project.json` → `packageDirectories[0]` (e.g., `force-app`). Read it from the project file before running commands.
-`--primary-field`: Most important identifying field (e.g., Name)
-`--secondary-fields`: Record summary (recommended 4-6, max 12)
-`--detail-fields`: Full record details, including required fields (e.g., Name)
**AppPage:**
- No additional requirements
**HomePage:**
- No additional requirements
#### **Field Selection Rules**
- **Validate fields exist**: Use MCP tools or describe commands to discover available fields for the object before specifying them in the command
- **Prefer compound fields**: Use `Name` (not `FirstName`/`LastName`), `BillingAddress` (not `BillingStreet`/`BillingCity`/`BillingState`), `MailingAddress`, etc. when available
- **Include required fields in detail-fields**: Always include object required fields (like `Name`) in the `--detail-fields` parameter, even if they're also used in `--primary-field` or `--secondary-fields`
After the base page deploys successfully, if the user wants additional components, follow the **Adding Components Dynamically** workflow below. All component additions MUST go through the discovery and inference pipeline — never write component XML from memory alone.
**Read `references/xml_rules.md`** for all XML encoding rules, field reference format, region/facet types, fieldInstance structure, unique identifier requirements, and common deployment error resolutions.
**Key rules (quick reminder):**
- Encode HTML in `<value>` tags: `&` first, then `<`, `>`, `"`, `'`
- Field references: `Record.{FieldApiName}` (never `Object.Field`)
- Every `<identifier>` and region `<name>` must be unique across the file
- Multiple components in same facet → combine in ONE region with multiple `<itemInstances>`
**Read `references/identifiers_and_regions.md`** for the identifier generation algorithm, facet naming patterns (named vs UUID), region selection rules, and container component facet structure.
> **MANDATORY WORKFLOW:** ALL component additions to a FlexiPage MUST follow this workflow. Do not write component XML from memory or skip discovery. This applies to standard OOTB components, custom LWC components, and any other component type.
**Key rule:** Discover ALL components as a batch first (one scan, one MCP call), THEN infer properties for each. Do not invoke discovery steps per-component or interleave discovery and inference.
After Tier 1 completes for all components, collect only the unresolved ones and pass them to Tier 2 in a single MCP call. Only components still unresolved after Tier 2 proceed to Tier 3.
The script scans `<packageDirectory>/**/lwc/*/`, tokenizes camelCase component names, scores them against user intent keywords, and returns a JSON array of matches with confidence tiers:
Run Tier 1 for ALL components before moving any to Tier 2. Collect all unresolved components from Tier 1, then pass them to Tier 2 in a single MCP call.
Call `getUiComponentSchemas` only when the component does **not** exist on the local machine (i.e., Tier 2 org-discovered components). For local components (Tier 1), extract `@api` properties directly from the source code instead.
If a file is returned: read and follow its inference rules, XML patterns, and defaults.
If empty: skip directly to Step 3.
Instructions files augment the schema — they provide *how* to derive values from user intent. Any properties in the schema not covered by the instructions file are resolved in Step 3.
### Step 3: Resolve Remaining Properties
For any properties not yet resolved, apply in this priority order:
**3a. Smart Default Heuristics:**
| Property Pattern | Default Value |
|-----------------|---------------|
| `recordId` | `{!recordId}` |
| `objectApiName` / `sObjectName` | Page's `<sobjectType>` value |
| `show*` / `visible*` / `display*` | `true` |
| `hide*` / `hidden*` / `disabled*` | `false` |
| Boolean without prefix | `false` |
| Schema specifies `"default"` | Use schema default |
**3b. LLM Inference:**
Use the component schema + user intent + page context to infer reasonable values. Example: user says "report showing top opportunities" → infer `reportName` should reference an opportunities report.
**3c. User Prompts (last resort):**
Only prompt the user for critical required properties that cannot be inferred. If user says "skip", omit the component entirely.
**To add a new component pattern:** Create `references/<namespace>_<componentName>.md` following the structure in existing files. The skill automatically checks for the file during Step 2 of property inference.
