v0.0.79

Clearer page limits, more armor for audits of slow or messy sites, and a lot of under-the-hood groundwork for auditing much bigger sites. More on that soon.

Added

• Audits tell you when they hit your plan's page limit. If a site has more pages than your plan covers in one cloud audit, the report and your audit listings now say the crawl was capped and at what count, instead of quietly auditing a subset. Coding agents get the same notice through the MCP server, so your agent knows it saw part of the site, not all of it.
• squirrel crawl gets the speed controls. The --concurrency and --per-host flags and the localhost fast path that squirrel audit gained in v0.0.74 now work on the standalone crawl command too.

Fixed

• Very slow or bot-hostile sites can no longer stall a cloud audit. Analysis of each asset and network probe now runs under its own time budget, and sites that drip-feed bytes to waste crawler time are detected and cut off. An audit of a hostile site finishes with what it could get instead of hanging until the overall timeout.
• More publish hardening for messy sites. Absurdly long URLs are skipped during the crawl instead of gumming up the works later, every remaining free-text field is trimmed at publish time, and size limits are now enforced byte-accurately so emoji and other multi-byte characters can't split at a boundary. This closes out the class of failures where one extreme page could sink an otherwise good report.

A reliability release for cloud audits. Heavy sites that used to kill the audit engine outright now complete, the easiest way for a good audit to die at the finish line is gone, failed audits always refund, and audits show you they are alive while they work.

Fixed

• Audits of heavy sites complete instead of dying silently. Sites with very large pages could exhaust the audit engine's memory mid-analysis: the audit went quiet and eventually showed up as timed out, with no report after multiple retries. Audit engines now run with 16x the memory and 8x the CPU, and the exact audits that failed this way now run to completion. If the engine ever does die, the audit is marked failed within moments, with a clear message and an automatic refund, instead of hanging in silence.
• Audits no longer fail because a page has very long text. A single oversized headline, or any absurdly long string anywhere on a site, could fail the whole audit at publish time: no report, just a refund and a shrug. Publishing now trims display text to fit instead of rejecting the report, across every field, and the same guard covers long lists (a site with issues on more than a thousand pages could trip it too). Big messy sites are exactly the ones that need auditing, so this one mattered.
• Blocked-site audits refund automatically. When a site blocks the crawler outright (bot protection, 403s at the door), the audit failed but kept its base charge, because the failure notice it published counted as a delivered report. Refunds now check what the report actually says. Every failure path, blocked, down, errored, cancelled, or stalled, nets zero credits.
• Post-crawl steps have hard deadlines. One audit sat for 52 minutes in a stuck enrichment call after its crawl had already finished. Every post-crawl step now has a deadline: a hung enrichment is dropped, its section is omitted, and the audit finishes with everything else intact. An overall backstop bounds the whole run even if a future step misbehaves.
• Deleted websites stay deleted. If a bookkeeping step failed during a delete, the site could keep showing up in your website listings until you deleted it again. The record now retries durably until it lands.

Added

• Live heartbeat while an audit works. The audit feed now shows "still working" progress every 30 seconds through the slow parts, so a long analysis is visibly alive instead of frozen, and a genuinely stuck audit is caught and refunded much sooner.