hot-updater-agent

HotUpdater Agent

Use this skill from the HotUpdater repository root when E2E should run through the dashboard queue or manual QA needs prepared provider infrastructure.

The CLI infers the current PR with gh pr view; there is no -pr flag.

Profiles

Valid profiles:

• standalone-dynamodb
• standalone-drizzle
• standalone-prisma
• standalone-kysely
• standalone-mongodb
• supabase
• cloudflare
• firebase
• aws
• vercel

Use the user-mentioned profile. If none is mentioned, use standalone-kysely.

Commands

Queue and wait for the normal AI verification loop:

hot-updater-agent verify -platform <full|ios|android> -profile <profile> -env-target examples/v0.85.0/.env.hotupdater

Queue without waiting:

hot-updater-agent e2e -platform <full|ios|android> -profile <profile> -env-target examples/v0.85.0/.env.hotupdater

Inspect current PR jobs:

hot-updater-agent status -limit 5

Inspect recent successful baselines by profile:

hot-updater-agent status -latest-success-by-profile -limit 20
hot-updater-agent -json status -latest-success-by-profile -limit 20

Inspect an exact job:

hot-updater-agent wait <task-id> -tail 240
hot-updater-agent reason <task-id> -tail 240
hot-updater-agent log <task-id> -tail 240

Inspect provider and stage bottlenecks. Prefer profile lookup for provider diagnosis because it resolves to the latest successful job for that profile:

hot-updater-agent timeline <profile|task-id> -limit 10
hot-updater-agent -json timeline <profile|task-id> -limit 10

Prepare one platform for agent-device-led manual QA:

hot-updater-agent manual start <profile> \
  -platform <ios|android> \
  -ref "$(git rev-parse HEAD)" \
  -env-target examples/v0.85.0/.env.hotupdater \
  -ttl 2h

manual start waits until the profile service, native release artifact, and control server are ready. Human output returns all handoff fields and exact agent-device commands. Prefer structured output when another agent consumes the handoff:

hot-updater-agent -json manual start <profile> -platform ios -ref "$(git rev-parse HEAD)"

The JSON session contains id, status/expiry/ref/log metadata, and handoff.worktreePath, scenarioRoot, envTargetPath, controlBaseUrl, runtimeConfigUrl, appBaseUrl, cleanupCommand, plus handoff.agentDevice with session, platform, device ID, app ID, binary path, and exact boot/install/open/snapshot/close commands.

Inspect or release the lease:

hot-updater-agent manual status [session-id] [-limit 20]
hot-updater-agent manual stop <session-id>
hot-updater-agent manual --help

Only one manual session may be active. It blocks dashboard E2E execution until manual stop or TTL expiry. Always stop it in a finally-style cleanup path; iOS and Android manual sessions must run sequentially.

Workflow

1. Choose full E2E (verify) or prepared manual QA (manual start).
2. Check existing work: hot-updater-agent status -limit 5 and, for manual work, hot-updater-agent manual status -limit 5.
3. For provider diagnosis, read hot-updater-agent -json timeline <profile> -limit 10.
4. For a failed exact E2E job, read hot-updater-agent reason <task-id> -tail 240.
5. For manual QA, consume only the returned handoff and always stop the lease.
6. Re-run the same path until it succeeds or the blocker is clearly outside repository code.

Use timeline data to distinguish setup, deploy, service boot, app reload, and E2E command execution bottlenecks. Compare providerBottlenecks[*].totalMs, providerBottlenecks[*].slowestStage, and global bottlenecks.

Default suite contract

Full-platform verify reads e2e/scenario-names.json from the exact checked-out PR commit. Adding a scenario to that manifest automatically includes it in every full-platform verification profile. A missing, empty, malformed, or duplicate manifest is a verification failure; do not replace it with an agent-side fallback list.