Share a Playwright HTML report
Every team that runs Playwright in CI knows the ritual. A test fails on
main. Someone posts the CI link in Slack. Whoever clicks it logs into the
CI provider, finds the right workflow run, downloads playwright-report.zip,
unzips it, opens index.html locally — and then screenshots the failure
back into Slack anyway, because there's no way to point at it otherwise.
Three structural problems, none of them Playwright's fault:
- The report is buried. It lives inside an artifact zip, behind CI
login, N clicks from the Slack message that mentioned it.
- The link expires. CI providers retain artifacts on a schedule —
the run you want to reference from last month's incident review is
often already gone.
- There's nowhere to discuss it. A failing trace deserves a thread
anchored to that failure. Instead the conversation happens in Slack,
detached from the artifact, and evaporates into scrollback.
The fix is the same one-curl pattern Comma uses for coverage
and eval reports: publish the run's HTML to a stable URL, let reviewers
comment on the failures in place, and append a revision per run.
The pattern
Store a scoped token (just reports:write) in your CI
secret store. Create the report once, save its id, and PATCH it on every
run so there is one URL per suite, not one per run:
curl -fsS -X PATCH "https://commareports.com/api/v1/reports/$REPORT_ID" \
-H "Authorization: Bearer $COMMA_API_TOKEN" \
-H "Content-Type: application/json" \
-d "$(jq -n --rawfile html run-summary.html '{html: $html}')"
Each push appends a revision. Reviewers can diff any two runs — "which
tests changed state since the last green build?" is a diff, not a memory
exercise — and the bookmarked link never goes stale.
In GitHub Actions:
- name: Publish Playwright summary to Comma
if: always()
env:
COMMA_API_TOKEN: ${{ secrets.COMMA_API_TOKEN }}
REPORT_ID: ${{ vars.COMMA_E2E_REPORT_ID }}
run: |
curl -fsS -X PATCH "https://commareports.com/api/v1/reports/$REPORT_ID" \
-H "Authorization: Bearer $COMMA_API_TOKEN" \
-H "Content-Type: application/json" \
-d "$(jq -n --rawfile html run-summary.html '{html: $html}')"
The if: always() matters — the runs you most want to share are the red
ones. The same two lines work in GitLab CI, CircleCI, or Buildkite; it's
plain HTTPS.
The honest part: it depends on how big the run is
Comma stores report HTML verbatim and renders it inside a sandboxed
iframe (sandbox="allow-scripts", no allow-same-origin). Scripts run —
the frame just can't reach the app's DOM, cookies, or storage. So
Playwright's own report, which inlines its run data into index.html as
a base64 zip, publishes and stays interactive: filtering works, steps
expand.
Where it stops working is size and siblings. The report body is capped at
5 MB, and a few hundred tests with inline attachments clears that easily.
Screenshots, videos and traces live in playwright-report/data/; drop
the whole directory in and those relative src/href references are
rewritten to the uploaded assets, but a report that fetches its data over
XHR at view time has an opaque origin and nothing to resolve the path
against.
So: publish index.html directly when the suite is small. For everything
else — and for the review experience, which is better anyway:
- Publish a static digest as the report body. A plain HTML table of
tests, statuses, durations, and error text, generated from Playwright's
JSON reporter output (
--reporter=json) by a small script in your
pipeline. A ~30-line script covers it, and the result is exactly the
surface a reviewer wants to comment on. Failure screenshots can be
inlined as images.
- Attach the heavy artifacts as assets. Trace zips, videos, and the
full report archive go in as report assets — 25 MB per
file, 250 MB per report — so they live next to the discussion instead
of in an expiring CI artifact.
- Step through traces in the trace viewer. When a failure needs
frame-by-frame investigation, download the attached trace and open it
in Playwright's trace viewer. Comma is where the discussion lives, not
a trace debugger — and it doesn't pretend to be.
If you need the full interactive report preserved verbatim, a static host
plus the artifact zip still does that job. What it won't give you is the
next section.
Comments land on the failure, not near it
The published digest renders inside Comma's sandbox with an anchored
comment layer on top. A reviewer highlights the row for
checkout.spec.ts › applies discount code and pins a thread to it:
"selector changed in Tuesday's release — fix in #482." The thread stays
attached as revisions accumulate, so when the same test flakes again in
three weeks, the prior discussion is one click away instead of buried in
Slack.
The loop closes with agents, too: a Claude Code or Cursor agent attached
via Comma's MCP server can read those threads with
list_comments, fix the selector, and reply on the thread — same scoped
token as the CI publish.
Per-run hygiene
- One report per suite, one revision per run — PATCH, don't POST,
and title revisions by commit (
"E2E — a1b2c3d") so the history reads
like a log.
- Rate limits are a non-issue — per-token limits default to 60
requests/minute; one publish per build is nowhere near it.
- Announce red runs — add a webhook notification on
revision.created and the new revision posts itself to Slack or
Discord. The Slack message goes back to being a notification; the
report is the artifact.
- Keep it team-visible — CI-published reports follow the same
sharing model as everything else. E2E results usually
want
team or private, not public-by-default.
Try it
Comma is free — unlimited reports, unlimited commenters, unlimited
revision history. Take the next red run you would have zipped, publish
the digest, and send the link instead.
Create your first report →
Related