Diff Intake

Diff Intake

When to Use

• Starting an instrumentation workflow from a pull request URL or branch name
• Auditing what analytics-relevant changes exist in a feature branch before it ships
• Converting a vague feature description into a concrete list of trackable interactions
• Feeding downstream skills (discover-event-surfaces, instrument-events) with structured input
• Reviewing merged PRs retroactively to find missed instrumentation

Core Jobs

1. Classify Input Type

Determine what the user has provided and select the right ingestion path:

• PR URL (github.com/.../pull/123) → fetch the diff via GitHub API or gh pr diff
• Branch name (feat/checkout-v2) → run git diff main...feat/checkout-v2 to get full changeset
• File or directory path → read the files directly and look for new/changed code
• Feature description (free text) → search the codebase for relevant files using keywords

Output of classification: input_type + source_reference + list of files changed.

2. Parse Changes

For each changed file, extract:

• What was added (new functions, components, API routes, state variables)
• What was removed (deleted screens, deprecated flows, removed UI elements)
• What was modified (changed behavior in existing flows — renamed fields, reordered steps)
• New UI surfaces: screens, modals, drawers, tooltips, banners
• User interactions: buttons, forms, links, gestures, keyboard shortcuts
• Data flows: new API calls, changed request/response shapes, new data dependencies

Focus on user-facing changes only. Skip internal refactors with no UX impact.

3. Extract Analytics-Relevant Changes

Filter parsed changes down to what matters for analytics:

• New screens → candidate page_viewed events
• New buttons/CTAs → candidate button_clicked or action-specific events
• New forms → candidate form_started, form_submitted, form_abandoned events
• New API endpoints exposed to users → candidate conversion or action events
• State changes with business significance → candidate status-change events
• Removed features → events that may need deprecation
• Changed flows → events that may need property updates

4. Produce Structured YAML Brief

Output a YAML document consumed by discover-event-surfaces:

brief:
  summary: "One sentence describing the feature change"
  pr_reference: "<URL or branch>"
  files_changed: 12
  new_surfaces:
    - type: screen
      name: "Checkout Confirmation Page"
      file: "src/pages/checkout/confirmation.tsx"
      interactions:
        - "Continue Shopping button"
        - "View Order Details link"
    - type: form
      name: "Shipping Address Form"
      file: "src/components/ShippingForm.tsx"
      fields: ["street", "city", "zip", "country"]
  removed_surfaces:
    - type: screen
      name: "Legacy Cart Page"
      note: "Replaced by new unified checkout"
  changed_behaviors:
    - description: "Payment step now skipped for free orders"
      files: ["src/flows/checkout.ts"]
      impact: "high — affects conversion funnel"
  technical_context:
    stack: "React + TypeScript"
    analytics_sdk: "Amplitude Browser SDK v2"
    existing_tracking_file: "src/analytics/events.ts"

MCP Tools

This skill uses minimal MCP — primarily local code analysis tools:

• No mcp__Amplitude__* tools required at this stage
• The output YAML is the input to discover-event-surfaces which uses Amplitude tools
• If the input is a GitHub PR URL, use web fetch or GitHub CLI to retrieve the diff

Key Concepts

• Analytics brief: The YAML document produced by this skill — the contract between diff analysis and event discovery
• New surface: Any new user-facing UI element that a user can interact with or see
• Changed behavior: An existing flow that works differently — may require property updates on existing events
• Technical context: Stack, SDK, and conventions that downstream skills need to write correct tracking code
• Funnel-critical change: A change that adds or removes a step in acquisition, activation, or conversion — highest priority for tracking

Output Format

A structured YAML brief with these top-level sections:

brief:
  summary: <string>
  pr_reference: <URL or branch name>
  files_changed: <count>
  new_surfaces: <list of {type, name, file, interactions}>
  removed_surfaces: <list of {type, name, note}>
  changed_behaviors: <list of {description, files, impact}>
  technical_context: {stack, analytics_sdk, existing_tracking_file}

After producing the YAML, state: "Brief ready. Proceed to discover-event-surfaces to map these surfaces to candidate analytics events."