| name | docs-visual-review |
| description | Verify OpenScribe docs locally or on deployed Pages by running Playwright screenshots with optional uv build and serve steps. |
| metadata | {"short-description":"Visual docs review with Playwright"} |
Skill: docs-visual-review
Purpose
Run fast and repeatable visual QA for the docs site with local automation.
Use this when
- You need to verify docs layout or styling changes.
- You want deterministic screenshots for review.
- You want one command that builds, serves, captures, and tears down.
Invocation in agent sessions
- Invoke by skill name and optional params, for example:
$docs-visual-review --out artifacts/docs-visual/latest.
- The agent runs and verifies the workflow steps.
Run
From repo root:
zsh .agents/skills/docs-visual-review/scripts/run.sh
Include app build precheck:
zsh .agents/skills/docs-visual-review/scripts/run.sh --with-swift-build
Custom output path:
zsh .agents/skills/docs-visual-review/scripts/run.sh --out artifacts/docs-visual/latest
Verify deployed docs without local serve:
zsh .agents/skills/docs-visual-review/scripts/run.sh --remote-url https://openscribe.dev/ --out artifacts/docs-visual/remote-latest
Keep server running after capture:
zsh .agents/skills/docs-visual-review/scripts/run.sh --keep-server
Workflow
- Verify required tools (
uv, playwright-cli, curl, swift).
- Use repo-local uv cache under
.build/uv-cache.
- Local mode: build docs (
uv run mkdocs build --strict) unless skipped.
- Local mode: start docs server (
uv run mkdocs serve) unless --no-serve is set.
- Remote mode: use
--remote-url to skip local serve and docs build, then verify deployed pages directly.
- Open Playwright browser, navigate to key docs routes, and capture full-page screenshots.
- Write logs and a short report.
- Review generated screenshots with the image tool and summarize visual findings.
- Close Playwright and stop the docs server (unless
--keep-server is set).
- Optional precheck: run
swift build only when --with-swift-build is set.
Required verification output
- Always report both:
- Script status from
report.md
- Visual findings after inspecting
home.png, ui-reference.png, and product-spec.png
- Visual inspection method:
- Open and inspect the generated screenshots directly with the image tool.
- Do not run auxiliary image processing commands such as
ffmpeg, ImageMagick, pixel probes, or ad hoc dimension scripts unless Stefan explicitly asks for that deeper analysis.
- If visuals and report disagree, treat the run as failed and explain the mismatch.
Outputs
Default output directory:
- Relative to repo root:
artifacts/docs-visual/<timestamp>/
Artifacts:
swift-build.log
mkdocs-build.log
mkdocs-serve.log
playwright.log
home.png
ui-reference.png
product-spec.png
report.md