Telemetry categories: opt-in selection + lightweight default set

[Sayan-]

What changes

• Category selection is now opt-in. A browser config captures exactly the categories set to enabled: true; anything omitted is off (previously omitted = on). A request that enables nothing clears telemetry.
• The default set (no browser config given, e.g. enabled: true) is now the lightweight operational signals: control, connection, system, captcha. The CDP categories (console, network, page, interaction) and screenshot are dropped from the default and must be opted into.

Why

"Just turn it on" should not spin up the CDP collector or unleash the high-volume streams. With this default, the collector only starts when a CDP category is explicitly enabled, and monitor (collector health) rides along only then. This is the firehose-safe default.

Test plan

• Rewrote tests
• Updated telemetry session + collector-failure tests for the non-CDP default
• Full non-e2e suite passes

---

Note

^{Cursor Bugbot is generating a summary for commit 71d3c04. Configure here.}

[firetiger-agent]

Created a monitoring plan for this PR.

What this PR does: Changes browser telemetry to fully opt-in — the CDP categories (console, network, page, interaction) are removed from the default capture set so the CDP collector is never started unless explicitly requested.

Intended effect:

• Default telemetry volume: baseline was CDP categories emitted for all default-configured sessions; confirmed if console/network/page/interaction events drop toward zero for sessions without explicit browser config.
• PUT /telemetry 400/500 rate: baseline 0/hr; confirmed if it stays at 0 post-deploy (no regression in the config-apply path).
• API 5xx rate: baseline 0.001–0.077% during active hours; confirmed if it remains in this range post-deploy.

Risks:

• Silent telemetry data loss for partial configs — any caller sending a partial browser config (e.g., only screenshot: true) and expecting other categories from old defaults will silently receive nothing for those categories; alert if users report missing console/network/page events in sessions that previously captured them.
• PUT /telemetry 500s (CDP collector start failure) — new error in API logs: "failed to start telemetry" where none existed before; alert if >5/hr.
• POST /browsers 500 regression — if a VM image bug crashes on startup; alert if 500 rate >0.5% for 2+ consecutive hours (baseline 0.001–0.077%).
• New telemetry ERROR patterns — pre-existing "copy telemetry stream / context canceled" logs (5–17/hr) are expected noise; alert on any new error body text in telemetry-related log lines.

Status updates will be posted automatically on this PR as monitoring progresses.

View monitor