Codacy just dropped Verity into public beta—and if you've been watching the chaos unfold as AI coding agents flood repos with unvetted output, this is the gate you didn't know you needed. Built by Codacy (yes, the same team behind that code quality CLI you probably have in your pipeline), Verity is an adversarial review layer designed to catch and repair security gaps, bad code patterns, and intent misalignments before any agent can commit. The twist? It's self-healing—the model reviews every change and fixes what it finds, so the next agent run starts clean.

The Problem That's Getting Worse

The numbers are stark. In four years, AI's share of new code written climbed from 5% to 75%, while security review pass rates dropped below 55%. Meanwhile, AI-tool spending hit $10B in 2026 and keeps climbing. More volume, less trust, higher cost—that's the gap Verity is built to close. Without an independent gate on development, every agent ships unreviewed code straight downstream, which means your review stage overheats while entropy and technical debt pile up beneath it.

Three Pillars: Gate, Memory, Cost

Verity wraps around your repo with three core capabilities. First, the gate itself—another model that didn't write the code judges every change against your coding standard across quality, security, and intent using Codacy's deterministic analysis CLI. It passes, or it's blocked with file-and-line findings and fixes, and the agent self-heals. Second, memory—a Markdown knowledge base committed to your repo captures every decision, gotcha, and integration pattern, feeding forward so each session starts smarter than the last. Third, cost observability—spend tracked per model, per session, per run, down to a fleet cost tree across every agent in your org.

How It Works Today

Getting started is straightforward: npm install -g @codacy/verity-cli && verity init. Verity hooks into the agent stop event and gates before commit. Currently works with Claude Code, with OpenAI Codex support listed as "Soon" along with CI/CD pipelines, GitHub Agentic Workflows, and AI Gateway integrations on the roadmap. The review layer uses Codacy's established deterministic analysis CLI under the hood—legitimate tooling that's been in production for years.

Where Teams Get Stuck

Verity frames this around what they're calling the AI coding maturity scale. Phase 1 is autocomplete (Copilot-style). Phase 2 is prompting agents to implement tasks with human review during sessions—this is where most teams are today. Phase 3 is loop engineering, with autonomous agents running in parallel on the same task. The pitch: without a gate, Phase 2 and 3 become walls because review overheats and entropy compounds. Verity keeps the review stage cool by gating development pre-commit so entropy stays held and cost stays visible at every phase.

Pricing During Beta

Free. No credit card, no seat limits. Just install and start gating. Codacy says they'll give beta teams plenty of notice before introducing paid plans. Everything Verity does—independent AI plus deterministic review on every change, PASS/FAIL gate on the stop hook, security/quality/intent lenses, one shared versioned Standard per project, compounding knowledge base—is free during this period.

The Bottom Line

This is exactly the kind of independent control loop the ecosystem needs as we barrel toward agent-heavy development. Codacy's deterministic analysis pedigree gives Verity credibility that another LLM wrapper lacks. Free during beta means there's no excuse not to eval it on your Claude Code workflows—if you're running agents without a gate, you're just accumulating debt with extra steps.