Acceptance tests. Run twice. Match every time.
Runs the ten acceptance tests server-side against saved HTML fixtures. Each fixture is audited twice; the runs must agree on statuses and finding order.
10 acceptance tests (T01–T10)
Judgment checks call the language model, so a full run can take a minute or two.
Multi-page tests
Checks page discovery (sitemap, robots.txt, same-domain, sensitive-path and 25-page limits, link-crawl fallback) on local fixture sites, then audits three pages twice and confirms the site results and exports are reproducible.
5 multi-page tests (S01–S05)
Discovery rules on two fixture sites (sitemap and link-crawl), plus a reproducible 3-page site audit and exports.
Primary action verifier tests
Checks the read-only inspect_link tool with mocked responses: 200, 404, redirects, off-origin and private-address blocking, multiple candidates, script-only CTAs, timeouts and restrictions, informational pages, and hostile destination content. Each case runs twice and the trace must match the requests actually made.
10 primary action verifier tests (W2-01–W2-10)
All destination requests are mocked; no live link is requested. Judgment checks still call the language model.
Approval workflow tests
Checks the Week 3 proposal state machine: content vs technical findings, server-enforced approval gate, approval invalidation on edit, idempotent draft creation, timeout handling, non-draft status errors, missing credentials, hostile page content, unresolved claim detection, and new-post-only write (W3-01 through W3-12). WordPress API is mocked — no live writes.
12 approval workflow tests (W3-01–W3-12)
WordPress API is mocked — no live writes. Tests the state machine, server-enforced approval gate, idempotency, hostile content handling, and claim detection.