Test Core Flow

Test — Core Flow (orchestrator)

Runs the DNA Insights core test flow end-to-end by chaining sub-skills. Each sub-skill is also independently invocable — the registry defines which one to pick for which job.

Sub-skill registry

More sub-skills will be appended here as the flow grows (e.g. provider consult, result delivery). When the user asks for a specific step only, invoke that sub-skill directly instead of the whole flow.

Shared state — state/report.json

Path: .claude/skills/test-core-flow/state/report.json (relative to repo root).

• The orchestrator initializes/overwrites it at the start of a core-flow run. Sub-skills merge their results into steps.<skill-name> when they finish (whether run from the core flow or standalone).
• Written with the Write/Edit tools by Claude (browser scripts cannot touch the filesystem).

{
  "flow": "test-core-flow",
  "environment": "dev",
  "dashboard_url": "https://dev-dashboard.dnainsights.ai/#/",
  "started_at": "<ISO timestamp>",
  "steps": {
    "test-create-new-case": {
      "status": "done",
      "environment": "dev",
      "case_id": "CA-XXXXXXXX",
      "patient_id": "PT-XXXXXXXX",
      "medicare_id": "XXXXXXXXXXX",
      "patient": "Gkim test",
      "sections": { "patient_information": "saved", "opt_in_consent": "saved (No)", "primary_care_provider": "saved", "family_history": "skipped (locked)" },
      "case_status": "New",
      "screenshot": "case-submitted-fast-run.jpeg",
      "finished_at": "<ISO timestamp>"
    },
    "test-submit-case-to-pending": { "…": "…" }
  }
}

Cross-skill landmine — stuck "Loading… ⟳" section body

Applies to EVERY sub-skill (same dashboard): a case section's heading renders before its form body mounts, and the body can hang on Loading... ⟳ indefinitely. Waiting longer never resolves it — every input-level poll reads an EMPTY form, so the run aborts with a misleading message ("answers did not take effect", "radio never took the value", got: []).

Rule: body still "Loading…" after ~2s → click ANOTHER section, wait ~1s, click back. The remount loads instantly. Encoded as openSection() / clickSection() in fast-02, pss-02 and prov-02 (3 bounce attempts, then a clear body stuck on "Loading…" error). Never "fix" this by raising timeouts.

Environment rule (IMPORTANT)

• From /test-core-flow: the orchestrator asks the environment ONCE (Phase 0) and stores it in state/report.json. Sub-skills invoked by the flow MUST NOT ask the environment again — they read it from the report.
• Standalone: a sub-skill invoked on its own asks dev/prod itself, then still merges its results into state/report.json.
• Credentials are per environment. Every account block in the sub-skill data files is keyed by env — sale_account.dev / sale_account.prod (case-data.json), pss_account.dev / .prod (pss-data.json), provider_account.dev / .prod (provider-data.json). Pick the block that matches environment; dev accounts (@gkim.vn) do not exist on prod (@gkxim.com) and vice versa. Missing block for the chosen env → STOP and ask the user for that account.
• Prod caveats (not yet exercised): the patient auto-answer number 804-222-1111 and the provider display name DR VU Doctor are dev facts. On prod, verify both before relying on step 3.

Flow

Phase 0 — One question round

Ask in a SINGLE AskUserQuestion call (two questions):

1. Environment: dev (https://dev-dashboard.dnainsights.ai/#/) or prod (https://dashboard.dnainsights.ai/#/).
2. The patient's medicare_id (goes into Primary Policy # — needed by step 1).

Phase 1 — Init report

Write state/report.json with flow, environment, dashboard_url, started_at, empty steps.

Phase 2 — Run /test-create-new-case

Follow that skill's FAST MODE, skipping its environment question (use the report values). On success it merges its result block into steps["test-create-new-case"]. On failure: stop the flow, report which step and why.

Phase 3 — Run /test-submit-case-to-pending

Follow that skill's FAST MODE, skipping its environment AND case questions — case_id comes from steps["test-create-new-case"].case_id. On success it merges into steps["test-submit-case-to-pending"].

Phase 3.5 — Run /test-provider-call-flow

Follow that skill's flow, skipping its environment AND case questions — case_id comes from the report (case must be Pending). Requires the patient phone to be 804-222-1111 (set in case-data.json). On success it merges into steps["test-provider-call-flow"].

Resume rule

If a core-flow run failed mid-way, state/report.json keeps the partial state. On the next /test-core-flow invocation, check it first: when steps["test-create-new-case"].status == "done" and the user confirms they want to continue that run (offer via AskUserQuestion), skip Phase 2 and go straight to Phase 3 with the recorded case_id — do NOT create a duplicate case. Starting fresh overwrites the file.

Phase 4 — Final combined report

✅ test-core-flow done
   Environment: <dev|prod>
   1) create-new-case:        CA-… / PT-… — Draft → New ✅
   2) submit-case-to-pending: Compliance ✅ / Medications ✅ — New → Pending ✅
   Report: .claude/skills/test-core-flow/state/report.json

List any unresolved questions at the end, if any.