Browser Trace

Attach a second, read-only CDP client to a browser session that is already being driven by your main automation. The trace records the full DevTools firehose to NDJSON, polls for screenshots and DOM dumps in parallel, and slices everything into a directory tree that bash tools can search.

This skill does not drive pages — it only listens. Pair it with the browser skill, browse, Stagehand, Playwright, or anything else that speaks CDP.

When to use

• The user wants to debug a browser-automation run (failing form, missing element, hung navigation, JS exception).
• The user has a running automation and wants to attach a trace mid-flight without restarting it.
• The user wants to split a CDP firehose into network / console / DOM / page buckets.
• The user wants screenshots + DOM snapshots over time, joined to CDP events by timestamp.

If the user just wants to drive the browser, use the browser skill instead.

Setup check

node --version                                  # require Node 18+
which browse || npm install -g browse
which jq     || true                                # optional — used only for ad-hoc querying

Verify browse cdp exists:

browse --help | grep -q "^\s*cdp " || echo "browse cdp not available — update browse"

How it works

Every Chrome DevTools target accepts multiple concurrent CDP clients. Your main automation is one client; this skill adds a second one that only enables observation domains (Network, Console, Runtime, Log, Page) and never sends action commands.

The tracer has three pieces:

1. Firehose: browse cdp <target> streams every CDP event as one JSON object per line to cdp/raw.ndjson.
2. Sampler: a polling loop calls browse screenshot --cdp <target> --path <file> and browse get html body --cdp <target> on an interval (default 2s). The helper passes --cdp when it samples so it can attach to the traced target from its own process; once a browse daemon session is attached to a CDP target, follow-up commands in that session do not need to repeat --cdp.
3. Bisector: after the run, bisect-cdp.mjs walks raw.ndjson once, slices it into per-bucket JSONL files keyed by CDP method, and additionally bisects per page using top-level Page.frameNavigated events as boundaries.

Quickstart

Local Chrome

# 1. Launch Chrome with a debugger port (any user-data-dir keeps it isolated).
"/Applications/Google Chrome.app/Contents/MacOS/Google Chrome" \
  --remote-debugging-port=9222 \
  --user-data-dir=/tmp/chrome-o11y \
  about:blank &

# 2. Start the tracer.
node scripts/start-capture.mjs 9222 my-run

# 3. Run your main automation against port 9222.
browse open https://example.com --cdp 9222
# ...whatever the run does...

# 4. Stop and bisect.
node

Browserbase remote

Two helpers wrap the platform-side bookkeeping: bb-capture.mjs creates or attaches to a session and starts the tracer; bb-finalize.mjs pulls platform artifacts (final session metadata, server logs, downloads) into the run dir at the end.

Browserbase ends a session as soon as its last CDP client disconnects. Create with --keep-alive, then attach automation to the session’s connectUrl before or together with the tracer. bb-capture.mjs --new handles the keep-alive session and tracer setup; your automation still needs to attach.

export BROWSERBASE_API_KEY=...

# 1. Create a keep-alive session AND start the tracer in one step.
#    Prints the session id, connectUrl prefix, and a live debugger URL you
#    can open in a browser to watch the run interactively.
node scripts/bb-capture.mjs --new my-run

# 2. Drive automation. bb-capture stamped the session id into the manifest.
SID=$(jq -r .browserbase.session_id .o11y/my-run/manifest.json)
CONNECT_URL=

Attaching to a session that’s already running (e.g. one your production worker created) — bb-capture.mjs accepts a session id instead of --new:

# Pick a running session (filter client-side; browse cloud sessions list has no --status flag)
browse cloud sessions list | jq -r '.[] | select(.status == "RUNNING") | .id'

node scripts/bb-capture.mjs <session-id> mid-flight-debug
# ...tracer runs alongside the existing automation client; no disruption...
node scripts/stop-capture.mjs mid-flight-debug
node scripts/bisect-cdp.mjs mid-flight-debug
node scripts/bb-finalize.mjs mid-flight-debug   # without --release: leave the session running

What you get from the Browserbase platform

bb-capture.mjs adds a browserbase block to manifest.json (session id, project, region, started_at, expires_at, debugger URL). bb-finalize.mjs writes:

• <run>/browserbase/session.json — final browse cloud sessions get snapshot (proxyBytes, status, ended_at, viewport, …)
• <run>/browserbase/logs.json — browse cloud sessions logs output. Often empty. The CDP firehose in cdp/raw.ndjson is the source of truth; this is a side channel.
• <run>/browserbase/downloads.zip — files the session downloaded, if any (the script discards the empty 22-byte zip you get when there are none)

Session replay artifact fetching is deprecated and isn’t fetched. Use the screenshots + DOM dumps in screenshots/ and dom/ for visual ground truth.

The live debugger_url in the manifest opens an interactive Chrome DevTools view served by Browserbase — handy for watching a long-running automation while the tracer captures the firehose to disk.

Filesystem layout

.o11y/<run-id>/
  manifest.json                 run metadata: target, domains, started_at, stopped_at
  index.jsonl                   one line per sample: {ts, screenshot, dom, url}
  cdp/
    raw.ndjson                  full CDP firehose (one JSON object per line)
    summary.json                {sessionId, duration, totalEvents, pages[]} — see shape below
    network/{requests,responses,finished,failed,websocket}.jsonl   session-wide buckets (always written)
    console/{logs,exceptions}.jsonl
    runtime/all.jsonl
    log/entries.jsonl
    page/{navigations,lifecycle,frames,dialogs,all}.jsonl
    dom/all.jsonl                                                  (only if O11Y_DOMAINS includes DOM)

When a run was started via bb-capture.mjs, manifest.json also carries a top-level browserbase block: session_id, project_id, region, started_at, expires_at, keep_alive, debugger_url.

Summary shape

cdp/summary.json is the entry point for any analysis: it has session-level totals and a pages[] array indexed by top-level Page.frameNavigated. Per-page entries are emitted in navigation order (page 0 = first concrete URL).

{
  "sessionId": "45f28023-…",
  "duration": { "startMs": 1777312533000, "endMs": 1777312609000, "totalMs": 76000 },
  "totalEvents": 420,
  "pages": [
    {
      "pageId": 0,

startMs / endMs / durationMs are wall-clock ms, derived from manifest.started_at plus the offset of each event’s CDP monotonic timestamp. domains[*] only includes errors/warnings keys when non-zero.

Drilling in with query.mjs

For interactive exploration, use scripts/query.mjs <run-id> <command> instead of remembering paths:

node scripts/query.mjs my-run list                    # one-line table of pages
node scripts/query.mjs my-run page 1                  # full summary for page 1
node scripts/query.mjs my-run page 1 network/failed   # cat failed.jsonl for page 1
node scripts/query.mjs my-run errors                  # all errors across pages, attributed by pid
node scripts/query.mjs my-run errors

Behind the scenes it just reads cdp/summary.json and the cdp/pages/<pid>/ tree — feel free to bypass it with raw jq/rg once you know the shape.

Top traversal recipes

# All failed network requests (use jq -c to keep it line-delimited)
jq -c '.params' .o11y/<run>/cdp/network/failed.jsonl

# Find requests to a specific host
jq -c 'select(.params.request.url | test("api\\.example\\.com"))' \
  .o11y/<run>/cdp/network/requests.jsonl

# 4xx/5xx responses
jq -c 'select(.params.response.status >= 400)

See REFERENCE.md for the full jq recipe library and a method-by-method bisect map. See EXAMPLES.md for end-to-end debug scenarios.

Best practices

1. Use bb-capture.mjs on Browserbase: it enforces --keep-alive, fetches the connectUrl, captures the debugger URL, and stamps the manifest. Doing it manually invites mistakes.
2. Don’t --release a session you don’t own: bb-finalize.mjs --release is for sessions you created with --new. When attaching to a production session via bb-capture.mjs <session-id>, run bb-finalize.mjs without --release so the original automation keeps running.
3. Order matters for remote: on Browserbase, attach the main automation client before (or together with) the tracer, and create the session with --keep-alive. Otherwise the session ends as soon as the tracer’s WS closes.
4. Don’t poll faster than ~1s: each sample runs browser CLI read commands and screenshots Chrome. 2s is a good default.
5. Pick domains deliberately: defaults (Network Console Runtime Log Page) cover most debugging. Add DOM for DOM-tree mutations (very noisy) via O11Y_DOMAINS="$O11Y_DOMAINS DOM".
6. Reuse one Browserbase session for the automation client on remote by attaching to that session’s connectUrl with browse open ... --cdp "$CONNECT_URL" --session <name>. The --session flag names the local browse daemon; it is not a Browserbase session attach flag.
7. Always run stop-capture.mjs, even after a crash, so background processes don’t linger and the manifest gets stopped_at.
8. Bisect once per run: bisect-cdp.mjs is idempotent — it overwrites the per-bucket files from raw.ndjson each time.

Troubleshooting

• browse cdp exited immediately: usually means the target is unreachable (wrong port) or the Browserbase session has already ended. For remote, verify with browse cloud sessions get <id> — if status is COMPLETED, recreate with --keep-alive and attach automation first.
• Empty raw.ndjson even though processes are running: confirm a CDP client is actually driving the page. The tracer only emits events that the browser generates, so an idle browser produces ~5 lines of attach/discover messages and nothing else.
• Screenshots all look identical: check index.jsonl — if url doesn’t change, the page hasn’t navigated yet. The polling loop runs independently of the main automation’s pace.
• Browserbase session ends mid-run: it likely hit --timeout. Recreate with a higher timeout (BB_SESSION_TIMEOUT=1800 node scripts/bb-capture.mjs --new ...) or remove the timeout flag.
• bb-capture.mjs <id> says “not RUNNING”: the session you tried to attach to ended. List candidates with browse cloud sessions list | jq '.[] | select(.status == "RUNNING")' and try again.
• browserbase/logs.json is empty []: expected — browse cloud sessions logs is sparse in practice. The CDP firehose in cdp/raw.ndjson is the source of truth.
• Where’s the session recording (rrweb)?: session replay artifact fetching is deprecated; this skill doesn’t fetch it. Use the screenshot stream in screenshots/ and DOM dumps in dom/.

For full reference, see REFERENCE.md (opens in a new tab). For example debug runs, see EXAMPLES.md (opens in a new tab).