v0.0.175

Reports now say what happened to the local store and to cloud rendering: meta.storeIntegrity tells you when page bodies were evicted since an audit ran (JSON schema 1.10), and a render block records the render mode, the pages rendered and fetched plain, and why rendering fell back (schema 1.11), with a line on the console. squirrel audit and the MCP server share one run lifecycle, so the terminal status is sent once and a second Ctrl-C still lets it land, and a Ctrl-C'd crawl is recorded as stopped and resumes. links/broken-links counts the targets that answered a refusal or a throttle, ax/ai-crawlers warns when Googlebot is blocked for the audited URL, core/canonical flags a canonical that points at the home page, a11y/meta-refresh reads the Refresh response header, ax/content-signals tells absent, declared, invalid and not checked apart, and the cloud fetcher's waiter has one state instead of three booleans.

Added

• ax/content-signals tells absent, declared, invalid and not checked apart (details.state). A robots.txt that was refused, failed or cut at the 512 KiB read cap is reported as not checked instead of as no Content-Signal declared; a plain 404 still counts as absent.
• a11y/meta-refresh also reports a Refresh response header: a redirect names the target and delay, a header with no URL is a reload, and a page with both the header and a meta refresh gets one finding naming both. The rule now runs per page, and its solution recommends a 301 or 308 redirect.
• core/canonical flags a page whose canonical points at the home page. A template that hard-codes the canonical to the site root tells search engines that every page it renders duplicates the home page, and the pages drop out of the index; the URL is well formed, so the rule passed it. A new canonical-to-homepage check fails when a page that is not the home page declares the site root as its canonical (the tag first, then the Link header), and warns for a locale home page such as /en/. The root with a trailing slash, /index.html or the www or bare host is one root, a root with a query string is not, a page that redirects to the home page is not flagged, and every other page gets the same result as before. (#1685)
• links/broken-links counts the targets that answered a refusal or a throttle. Every broken-links row now carries refusedTargets (a 401, a 403 or a bot-wall challenge) and rateLimitedTargets (a 429) in its coverage details, next to linkTargets, probedTargets and unfetchedTargets. Those targets are not broken and not unfetched, and nothing is known of the page behind them, so linkTargets less unfetchedTargets over-counts what was observed; subtract these two as well. No check message, item or status changes. (#1679)
• A render that fell back to plain HTTP is visible in the report. When cloud rendering stops partway (out of credits, the render service failing, a signed-out or unreachable cloud), the audit now prints one line with the reason and how many pages were rendered, and the JSON report gets a render block (requested, mode, pagesRendered, pagesFetchedPlain and a fallbackReason) with a matching <render> element in the LLM report, so an agent can tell rendered pages from plain-HTTP ones. (#1664)
• squirrel pages and squirrel entities say when page bodies were evicted, and report puts the count in its JSON. Content-store eviction is global by size cap, so an audit's page bodies can be gone later, and only report, --resume and re-audits said so. pages and entities now print the same notice on stderr (their rows are built from stored page data, so the output is complete), and with --format json all three carry storeIntegrity (status: "evicted", evicted, stored): meta.storeIntegrity in the report JSON, a top-level field in pages and entities. It is absent when every body is still there. All readers share one census. (#1667)

Changed

• ax/ai-crawlers warns when Googlebot is blocked for the audited URL. The search-engine-access check was info only. A robots.txt that blocks Googlebot from the site root or for the entry URL usually means the site is not indexed, often by a staging Disallow: / left in, so that check now warns and counts toward the score. An Allow override, an allowed Googlebot and a block of any other engine (Bingbot, Applebot) stay info, and details.bots still lists every bot. (#1680)

Fixed

• A second Ctrl-C no longer leaves the run waiting for the server to reap it. squirrel audit handled the first interrupt with process.once, so a second Ctrl-C while the "cancelled" status was being sent killed the process under it, and the run sat pending until the server's reaper closed it as failed. The handler now stays installed: a later signal waits for the status already in flight, for at most 3 seconds, and then exits. The CLI and the MCP server also share one run lifecycle now, so the terminal status is sent at most once on every exit path and "running" can never be sent after it. (#1663)
• A Ctrl-C'd crawl is recorded as stopped, not left paused. The crawler's own stop() writes paused first, and the interrupt handler only marked crawls that were still running, so the stored status said paused where the audit was stopped. The handler now writes through the same guarded status write as every other change, which allows stopped from running or paused and never touches a crawl that already finished. --resume continues it as before. (#1672)
• A report's "failed or blocked" and --max-duration partial status each have one definition. isAuditFailed and isScoreless in @squirrelscan/core-contracts replace the check that was written out in about fifteen places, and one function, withMaxDurationCut, now makes a time-limited audit partial for both the live run and squirrel report rebuilding it from the store, so the two cannot disagree. Reports read as before. (#1662)