v0.0.52

squirrelscan v0.0.52

New

• Local MCP server — squirrel mcp runs squirrelscan over stdio so any MCP-compatible agent (Claude, etc.) can crawl and audit directly.
• Per-website crawl & audit config — set max pages, depth, and coverage per site, with trigger attribution on every run.
• Redesigned report header + one-click "open report" from the dashboard.

Cloud

• Smart audits — cross-audit finding merge with a union rescore, so a partial re-audit carries prior findings forward instead of inflating your score; removed findings are retired cleanly.
• More reliable cloud-render crawls — interruptible fetches, a crawl-phase watchdog, a stale-run reaper, and an HTTP-first hybrid mode for faster runs that no longer stall.

Reliability & fixes

• Publishing no longer fails when a site's own security findings look "code-like"; a failed publish is now recorded as an error audit with a clear status instead of vanishing silently.
• Website title/description are taken from the page's real metadata rather than an AI paraphrase.
• Existing reports pick up renderer changes when their HTML is regenerated.
• Crawler: quick-mode sitemap discovery is restored when resuming a crawl.

Performance

• Faster crawls — lower hot-path overhead and higher single-host throughput with politeness-aware staggering.
• Rules & reports — Core Web Vitals hints computed once per page, O(N²)→O(N) result accumulation, and batched result persistence.