Test All — Parallel Orchestrator

Runs all 9 IWSDK test suites simultaneously. The orchestrator handles the full lifecycle: build, example prep, dev servers, sub-agent launch, polling, retries, cleanup, and aggregate reporting.

Each test gets its own example directory and dev server. Sub-agents read the skill files and execute tests through the direct npx iwsdk flow described in each skill.

---

Test Map

Ports are not pre-assigned. Each dev server picks its own port dynamically. The orchestrator can discover each active URL from the server logs or npx iwsdk dev status, but the sub-agent skills no longer need the port explicitly.

---

Phase 1: Prerequisites & Build

node -v

pnpm -v

Verify node >= 20.19.0 and pnpm is available. Stop on failure.

pnpm install

cd packages/vite-plugin-dev && npx playwright install chromium && cd ../..

pnpm build:tgz

All must succeed. Stop on failure. Note: npx playwright must run from packages/vite-plugin-dev where playwright is a dependency.

---

Phase 2: Prepare Examples

Clone poke for tests that share it

5 tests use the poke example. Clone it into separate directories so each gets its own dev server:

node scripts/test-prep.mjs clone

Fresh install all 9 examples in parallel

node scripts/test-prep.mjs install

---

Phase 3: Start 9 Dev Servers

Start all dev servers, wait for them to be ready, and discover their active URLs. This single command handles everything — starting servers, polling for readiness, and outputting the current port map:

node scripts/test-servers.mjs start

The output is JSON with the current port map: {"poke": 8084, "poke-ecs": 8082, ...}. Keep it for diagnostics, but the sub-agents should rely on the direct CLI flow inside each example directory rather than passing ports around.

If any server fails to start within 60 seconds, the script reports which ones are missing and exits with code 1. Check /tmp/iwsdk-dev-<name>.log for errors.

To re-check ports later without restarting:

node scripts/test-servers.mjs ports

---

Phase 4: Launch 9 Sub-Agents

For each test, launch a Task sub-agent with subagent_type: "Bash", mode: "bypassPermissions", and run_in_background: true.

Sub-agent prompt template

Each sub-agent gets this prompt (with <SKILL_FILE>, <EXAMPLE_DIR>, and <ROOT> substituted):

Read the file at <ROOT>/.claude/skills/<SKILL_FILE> and execute the test instructions
starting from Step 3 (Verify Connectivity). Steps 1 and 2 have already been completed —
the dev server for <EXAMPLE_DIR> is already running.

Execute all test suites (Steps 3-4), then do Step 5 (cleanup and results summary).
Do NOT kill the dev server in cleanup — the parent agent will handle that.

Launch all 9 agents

Launch all 9 Task agents in a single message with multiple Task tool calls. This starts them concurrently.

…