Run exhaustive multi-agent code and app quality audit

Running a complete, adversarially-verified code + app quality audit without making edits; pair with quality-remediate workflow to apply fixes.

• · date (string, YYYYMMDD or labeled)
• · depth: quick|standard|exhaustive (controls scan scope + skeptic fan-out)
• · includeSim: boolean (capture live simulator screenshots)
• · scope: optional paths array to restrict search surface
• · autoRemediate: boolean (false by default)

• · REPORT.md (synthesized findings + verification verdicts)
• · findings.json (structured array: dimension, severity, file, line, title, description, evidence, suggestedFix, fixTier, knownTracked)
• · simulator screenshots (if includeSim=true)
• · per-dimension finder results (unverified candidates beyond DEPTH_CFG.maxVerify cap)

• · subagent fan-out (15 quality dimensions × ~13 modules × depth multiplier)
• · adversarial skeptics (per finding)
• · simctl (if includeSim=true)
• · vision analysis (simulator analysis)

• · read-only mode (no edits performed)
• · ~15 quality dimensions must be defined and scannable
• · DEPTH_CFG knobs tuned per depth (maxRounds, dryStop, skeptics, modules, critic, maxVerify)
• · rate-limiting guardrails in place (maxVerify hard-cap)

• · fan-out rate-limiting if maxVerify too high (2026-06-03: 808 agents, 47% throttled)
• · duplicated findings across dimension agents (requires post-deduplicate)
• · simulator capture fails if device is unavailable
• · skeptics timeout or produce inconclusive verdicts
• · scope filter silently ignores modules outside paths

• · 6-phase structure: Recon → Discover → Verify → Sim → Critic → Synthesize
• · 3 adversarial skeptics per finding (refute-by-default)
• · fixed severity enum: critical|high|medium|low|feature
• · fixed tier enum: mechanical|logic|clinical|architectural
• · DEPTH_CFG table with known rate-limit tuning (quick:25, standard:60, exhaustive:120 maxVerify)
• · per-finding knownTracked boolean (integrates with existing issue tracking)
• · evidence field captures offending code snippet or grep hit (verifiable)