| name | state-of-the-web |
| description | Run a web-uplift audit of the top 1000 sites and produce structured results.
Collects CDP evidence (screenshots, Lighthouse, axe, heap, layout, discoverability)
and judges 17 quality principles per site. Outputs per-site JSON + merged SQLite DB.
|
State of the Web — batch audit skill
When to use
When asked to audit, benchmark, or measure the quality of websites at scale. Produces structured per-site results suitable for aggregation and trend analysis.
Setup
node ~/.web-uplift/evidence/cli.mjs --help
pip install tldextract
curl -sS "https://tranco-list.eu/top-1m.csv.zip" -o /tmp/tranco.zip && unzip -o /tmp/tranco.zip -d /tmp/
Running
CDP evidence pass (automated, no vision)
python3 scripts/audit_runner2.py <site-list.txt> <start-index> <count>
- Outputs
results/cdp/results-batch-{start}.json
- ~10-20s per site
- Collects: CLS, overflow, JS-shell detection, discoverability, screenshots
Vision-based principle audit
See AGENTS.md for the full per-site methodology. Key steps per site:
- Gather evidence (screenshot, layout, discoverability, HAR, trace, heap, axe, probes)
- Review screenshots + metrics
- Judge all 17 principles (pass/issues/n/a with findings)
- Score the site (0-100)
- Save to
results/gpt/{site}.json
Merge + build SQLite
python3 scripts/merge_results.py
Output
results/cdp/ — CDP evidence (automated metrics)
results/gpt/ — Principle judgments (vision-based)
state-of-the-web.db — Merged SQLite database
evidence/ — Screenshots and artifacts (not in git)
Principles (17)
See AGENTS.md for the full list with descriptions and detection hints.
Integration with web-uplift
This skill uses the web-uplift evidence primitives as its data-gathering layer. The web-uplift skill defines the 17 principles and the evidence CLI; this skill adds the batch runner, SQLite schema, and aggregation.