layerkit-discover-data-layer

layerkit-discover-data-layer

Analyze the customer data layer (not vendor docs). Output is a domain_spec proposal that seeds Q3–Q4.

Protocol

1. List candidate files with narrow deterministic tools (rg --files, exact symbol search), then read source yourself. Do not use deterministic heuristics to infer business domains.
2. Scan TS/JS/Java (and similar) as the AI agent:
   • event / intent type names
   • analytics or vendor emit call sites
   • DTOs / domain models for user, cart, purchase
3. Deny-paths (never open or paste):
   • .env, .env.*
   • **/*secret*, **/*credential*
   • **/id_rsa*, **/*.pem, **/keystore*, **/*.p12
4. Prefer in-repo OpenAPI, JSON Schema, or redacted curl for field hints (source: openapi|curl|code).
5. Draft domain_spec proposal shape:

{
  "schemaVersion": 2,
  "kind": "domain_spec",
  "id": "domain-spec-<customer>-v1",
  "summary": "Customer domain intents/fields from code",
  "authoredBy": "agent",
  "status": "draft",
  "createdAt": "<ISO>",
  "sources": [
    { "title": "<file role>", "url": "file://<repo-relative-path>", "excerpt": "<short>" }
  ],
  "payload": {
    "id": "<customer-or-commerce>",
    "version": "0.1.0",
    "description": "...",
    "intents": [{ "id": "purchase", "description": "..." }],
    "fields": [
      { "path": "user.email", "type": "string", "description": "...", "required": false }
    ]
  },
  "maker": { "type": "agent", "id": "<agent>" }
}

1. Bootstrap questionnaire Q3 (intents) / Q4 (fields) only when code supports them; residual gaps stay unanswered (source: unanswered).
2. Persist:

layerkit proposal validate ./domain-spec.json
layerkit memory append --type research --title "domain discovery" --vendor general --body-file ./discovery-note.md

1. Next: layerkit-research-vendor (per target vendor).

Multi-vendor packages

When the client package already integrates multiple vendors (adapters, mappers, gateways, destinations):