Log in

update-screenshots

Vercel
All-time installs
2,661

Update the committed blocks-ci screenshot hashes after the "Screenshots & Tests" check fails, or investigate a screenshot diff reported on a PR. Use when asked to update, accept, or refresh component screenshot baselines from CI. This skill should be run as a subagent.

Other options

Summary

Update the committed blocks-ci screenshot hashes after the "Screenshots & Tests" check fails, or investigate a screenshot diff reported on a PR. Use when asked to update, accept, or refresh component screenshot baselines from CI. This skill should be run as a subagent.

Raw SKILL.md

5,075 bytes
---
name: update-screenshots
description: Update the committed blocks-ci screenshot hashes after the "Screenshots & Tests" check fails, or investigate a screenshot diff reported on a PR. Use when asked to update, accept, or refresh component screenshot baselines from CI. This skill should be run as a subagent.
---

# Update Component Screenshots from CI

Screenshot **images** are not stored in the repository — they live in an external service
(`hediet-screenshots.azurewebsites.net`), keyed by commit SHA. But a subset of fixtures is
pinned by **hash** in [`test/componentFixtures/blocks-ci-screenshots.md`](../../../test/componentFixtures/blocks-ci-screenshots.md),
and that file **is** committed. When those hashes change, CI fails and you must update the file.

## Two different outcomes, only one of which blocks

The `Screenshots & Tests` job in [`.github/workflows/component-fixtures.yml`](../../workflows/component-fixtures.yml)
produces two independent results:

| Result | Blocking? | Action |
| --- | --- | --- |
| Screenshot **diff report** (PR comment with before/after images) | No — informational | Review the visuals. Nothing to commit. |
| **blocks-ci hash mismatch** | **Yes — fails the check** | Update `blocks-ci-screenshots.md` and commit. |

A fixture opts into the blocking gate with `labels: { kind: 'screenshot', blocksCi: true }`.
Only those fixtures appear in `blocks-ci-screenshots.md`.

The failure looks like this:

```
##[error]blocks-ci screenshot hashes do not match committed file. See PR comment or job summary for the updated content.
```

## Step 1: Get the expected hashes from CI

> **Never regenerate the hashes locally.** They are hashes of the rendered PNG bytes, produced
> on `ubuntu-latest`. Rendering on macOS or Windows yields different bytes and therefore
> different hashes, so locally generated values will fail CI. Always copy the values from the
> CI job.

Three surfaces carry the same content — use whichever is handy:

- The **PR comment** titled "blocks-ci screenshots changed" (non-fork PRs only) — contains the
  full updated file plus a patch.
- The **job summary**, which gets the identical body and is the only surface fork PRs receive.
- The **job log**, whose final step prints a unified diff:

```bash
gh api repos/microsoft/vscode/actions/jobs/<JOB_ID>/logs > "$TMPDIR/ci-job-log.txt"
grep -n '##\[error\]' "$TMPDIR/ci-job-log.txt"
```

Find the failed job id with:

```bash
gh pr checks <PR> --json name,link,bucket --jq '.[] | select(.name == "Screenshots & Tests")'
```

## Step 2: Verify the change is intentional before accepting it

This gate exists to catch **unintended** layout regressions, so accepting new hashes without
looking at the images defeats its purpose. The images are publicly fetchable by hash, so pull
both the old (committed) and new (from CI) versions and compare:

```bash
curl -sL -o old.png "https://hediet-screenshots.azurewebsites.net/images/<OLD_HASH>"
curl -sL -o new.png "https://hediet-screenshots.azurewebsites.net/images/<NEW_HASH>"
```

Then view them, and localize the change rather than eyeballing full screenshots — the delta is
often only a pixel or two:

```bash
python3 -c "
from PIL import Image, ImageChops
a = Image.open('old.png').convert('RGB'); b = Image.open('new.png').convert('RGB')
print('diff bbox:', ImageChops.difference(a, b).getbbox())
"
```

Confirm the delta matches what the PR intends. If the fixture is unrelated to the change, or
the shift is larger than expected, treat it as a regression and fix the code instead of the
hashes.

## Step 3: Apply and commit

Edit only the changed lines in `test/componentFixtures/blocks-ci-screenshots.md`, replacing the
old hash in the image URL with the new one:

```md
#### editor/inlineChatZoneWidget/InlineChatZoneWidget/Dark
![screenshot](https://hediet-screenshots.azurewebsites.net/images/<NEW_HASH>)
```

The file is generated by [`build/lib/screenshotBlocksCi.ts`](../../../build/lib/screenshotBlocksCi.ts)
and compared **byte-for-byte**, so keep the `<!-- auto-generated by CI — do not edit manually -->`
header, the `#### <fixtureId>` / image-link pairing, the blank line between entries, and the
`fixtureId` sort order intact. Verify your edit is the exact inverse of the diff CI reported:

```bash
git diff test/componentFixtures/blocks-ci-screenshots.md
```

Then commit and push. The check re-runs and should pass; hashes on `main` become the new
baseline after merge.

## Investigating further

Raw captured images and the manifest for a run are uploaded as an artifact:

```bash
gh run download <RUN_ID> --name screenshots --dir .tmp/screenshots
```

`manifest.json` maps each `fixtureId` to its `imageHash` and any render errors.

## Related failures from the same job

The check also fails if a fixture **failed to render** (`Fail if fixtures had errors`) or if the
Playwright fixture tests failed. Those are genuine bugs — updating hashes will not help. Look
for `::error::<fixtureId>:` in the log, and download the `playwright-test-results` artifact for
test failures.

Security audits

SnykPASS
SocketPASS
Gen Agent Trust HubPASS