Detectors & audits

Where collectors measure, detectors judge. Each detector watches for one class of problem — a bloated DOM, a growing heap, a duplicated request — and emits a VibeIssue when a threshold trips. Thirteen ship enabled by default.

The Detector interface

A detector is a factory function create*Detector() that returns an object implementing:

interface Detector {
  readonly name: DetectorName
  start(): void
  stop(): void
  getIssues(): readonly VibeIssue[]
  clear(): void
}

The engine calls start() on each enabled detector, then reads getIssues() from all of them every snapshot. clear() empties a detector's issue buffer.

Issues carry evidence

Detectors create issues through a shared createIssue() helper, so every issue has the same shape — id, detector, severity, title, description, evidence, timestamp, plus acknowledged / resolved flags. The evidence object is typed per detector (see IssueEvidenceMap in the protocol), so a dom-bloat issue carries { nodeCount, maxDepth, selector } while a duplicate-requests issue carries { url, method, count, windowMs }. That evidence is what your agent reads to write a targeted fix.

The thirteen detectors

Ten catch runtime and payload problems; three are audits that emit one issue per failed check.

See the Detector & audit reference for every threshold and check id, each cross-linked to a /fix guide.

Toggling detectors

All thirteen are enabled by default. Turn any of them off through the engine's detectors config:

import { VibeCheckEngine } from '@wcgw/vibe-check-core'

const engine = new VibeCheckEngine({
  detectors: {
    consoleSpam: false, // quiet a chatty dev build
    seo: false,
  },
})
engine.start()