the adversarial audit

What the audit actually found.

"80 agents, 14 facets, zero criticals" — that's the headline. Here's what's behind it: every finding was refuted by a separate agent before it counted, then fixed and redeployed. Three of the real ones:

three of the real ones

HIGH Import CPU-DoS

The website importer's JSON-LD scan went quadratic on a page of unclosed <script> tags — a crafted 2 MB page froze the single-worker gateway for minutes. Fixed: a bounded, linear scan — 70s → 4ms.

HIGH Crisis behind the meter

The billing check ran before the crisis check, so a distressed customer messaging an out-of-credits business got "unavailable" instead of a crisis line. Fixed: the crisis check now runs before billing on every surface, served free.

HIGH Render DoS

A per-tenant social-card render (Pillow) re-ran on every cache-busting request. Fixed: memoized by content, so a URL loop can't re-trigger it.

14 facets: isolation · auth · injection · money · governance · concierge · booking · channels · customisation · site-generator · frontend · ops · layout · data. Verdict: zero critical; the XSS surface and tenant isolation came back clean; no live auth bypass — every confirmed edge fixed.

Re-run the audit's core checks live →

the honest numbers

Every number, with the asterisk already attached.

Most builders pad. I'd rather you trust the parts that are real than be impressed by parts that aren't.

1builder* not a team

819tests green* not a formal proof

80 → 0adversarial agents, zero criticals* self-run, not third-party

1production instance* by choice — I haven't had to learn scale yet

2 → 1residents, one spine* composition, proven by lint