feat(devtui): track usage telemetry for setup flows and dashboard

[Soner (shyim)]

Summary

Adds anonymous usage telemetry to the interactive dev TUI (shopware-cli project dev) so we can answer: where do the setup wizards lose users, do Docker environments start reliably, and which dashboard features are actually used.

All events go through the existing internal/tracking pipeline (pseudonymous ID, UDP fire-and-forget, DO_NOT_TRACK opt-out) and are fully documented in docs/TELEMETRY.md. Privacy posture is unchanged: no URLs, no sales-channel/theme names, no credentials, no free-text input — only enumerated choices, outcomes, and durations.

Events

shopware_cli.project.dev.install — first-run Shopware installation

Sent when the "Shopware is not initialized yet" wizard reaches a terminal state. Choice tags only appear once the user made that choice.

Tag	Values
result	success / failure / cancelled / skipped
abandoned_at	ask / language / currency / credentials / installing (cancelled only)
failed_step	last deployment-helper step that had started, e.g. system:install (failure only)
duration_ms	install runtime once the install actually started
language / currency	selected defaults, e.g. de-DE / EUR
custom_credentials	true/false — whether the default admin user/password were changed; the values themselves are never sent

shopware_cli.project.dev.migration_wizard — dev-environment setup wizard

Replaces the old migration_wizard_completed event, which only reported successful runs, so no conversion rate could be computed.

Tag	Values
result	completed / cancelled / failed
abandoned_at	welcome / admin_user / docker_php / review (cancelled only)
duration_ms	time since the welcome screen was confirmed
php_version	selected PHP version (completed/failed)
deployment_helper_added	whether composer.json was missing shopware/deployment-helper (completed only)

shopware_cli.project.dev.docker_start — container startup

Sent when a docker compose up -d run by the TUI finishes (or the user gives up waiting).

Tag	Values
trigger	initial / config_change
result	success / failure / cancelled (user quit while waiting)
duration_ms	startup time

shopware_cli.project.dev.action — dashboard actions

Sent per command-palette action. Instant actions carry only the name; task actions also report outcome and runtime.

Tag	Values
action	open-shop / open-admin / cache-clear / admin-build / sf-build
result	success / failure / cancelled (task actions only)
duration_ms	task runtime (task actions only)

shopware_cli.project.dev.watcher — admin/storefront watchers

Sent once per watcher run, when it ends. Crash rate vs. user-stop rate tells us if the watchers are reliable; uptime tells us if people develop inside the TUI or just use it as a launcher.

Tag	Values
watcher	admin / storefront
result	prep_failed / crashed / user_stopped / session_end
uptime_ms	how long the watcher ran

shopware_cli.project.dev.health — setup health snapshot

Sent once per session when the Overview tab's setup-health report first loads: one tag per check with its level as value. Shows how common misconfigurations are in the wild — and which checks are worth turning into automatic fixes. Only the level is sent, never the underlying values.

Tag	Values
php_version	ok / warn / critical
memory_limit	ok / warn
admin_worker	ok / warn
flow_builder_log_level	ok / warn

shopware_cli.project.dev.session — dashboard session shape

Sent when the user leaves the dashboard. Total command duration is already covered by shopware_cli.command; this adds what happened inside the TUI.

Tag	Values
executor	docker / local
duration_ms	TUI session length
tabs_visited	sorted, comma-separated, e.g. config,instance,overview
actions	number of palette actions invoked
watchers_used	admin,storefront (omitted if none)
exit	stop_containers / keep_running / quit

Implementation notes

• All bookkeeping lives in a new telemetryState, held by pointer on Model so Bubble Tea's value copies share it; every method is nil-safe because tests construct Model{} directly.
• Outcome events latch (installOnce, healthOnce, sessionTags, per-run watcher start times), so overlapping quit paths — e.g. pressing q on the install-failure screen, or a trailing logDoneMsg after a user stop — cannot double-report.
• Events on quit paths send synchronously with a 300 ms cap (a goroutine would race process exit); everything else sends async so the UI never blocks on DNS.
• A package TestMain sets DO_NOT_TRACK, so the devtui test suite can never emit real events.
• docs/TELEMETRY.md is updated with all events (and now also mentions the previously undocumented migration-wizard tracking).

Test plan

• go test ./... — new unit tests cover tag building (choice omission, credential redaction, once-latching, watcher double-report protection, health flattening)
• go vet, gofmt, go build

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7ba3e2e401

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "Codex (@codex) review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "Codex (@codex) address that feedback".

[Paul Riga (RigaPaul)]

I would extract all the string identifiers for events into constants

[Tomasz Turkowski (tturkowski)]

LGTM