Products
Quality and security for software built with AI.
AI tools write code fast, but the mistakes they make are quiet: a summary that doesn't match what ran, a destructive command that sounds reasonable, a missing authorization check, a file that has grown to do five unrelated things. Procli is five focused products that catch these before they reach production — two that watch your agent live, two dashboards that scan your repository, and a CLI that runs the same checks from the terminal.
The five products
Procli Cheat
Claim verification · Live
AI agents write confident summaries of things they never verified. Cheat treats every “tests pass,” “build succeeds,” “file created,” or “no vulnerabilities” as a claim, then checks it against what actually ran in the session — the commands executed, the files that changed, and the results those commands returned.
What it catches: Claims with no matching command, fixes to files that were never touched, partial test runs reported as full, and PR summaries that overstate what was verified. Each finding is labeled Confirmed, Unconfirmed, Partial, or Contradicted.
Works with: Claude Code, Terminal, GitHub
Procli Guard
Agent guardrails · Live
Before a tool call executes, Guard reads the action and weighs how reversible it is and how wide its blast radius is — separately from how confident the agent sounds. Clearly destructive actions are blocked automatically, low-risk work keeps moving, and the ambiguous middle is flagged for you to decide on.
What it catches: Destructive shell commands, accidental deletions, unverified package installs, secrets files being read, auth or payment logic changed without review, and instructions smuggled in through issues, comments, or logs.
Works with: Claude Code, Terminal, GitHub
Procli Quality Dashboard
Code quality · Dashboards
Quality figures out what each file actually is — a component, an API route, a service, a config file, a test — and judges it against what that role calls for, then scores structure, complexity, duplication, and overall project health. You connect a repository from GitHub, click run scan, and get prioritized findings with fixes, plus JSON you can hand straight to your agent.
What it catches: Oversized files and functions, duplicated logic that should be centralized, business logic leaking into UI components, mixed responsibilities in one file, and orphaned files or risky circular dependencies.
Works with: GitHub, repositories, dashboards, teams
Procli Audit Dashboard
Security review · Dashboards
Audit scans your existing codebase for a specific pairing — a risky pattern present alongside the safety check that should sit next to it but doesn't — and flags that combination wherever it appears. Findings map to files, vulnerability classes, and severity in one dashboard built for action, with fixes and JSON output.
What it catches: Hardcoded or publicly exposed secrets, private routes with no authentication, user data accessed without an ownership check, uploads with no type or size validation, and raw internal errors returned straight to users.
Works with: GitHub, repositories, security review, teams
Procli CLI
Terminal workflow · Local
The CLI runs the same Quality and Audit checks from the terminal, before work is pushed or shipped. Run procli login once to link the CLI to your account with a short device code, then run checks locally — the check and its output are identical no matter which agent wrote the code.
What it catches: procli audit and procli quality print a clear report with exact files and line numbers, plain-language explanations, severity levels, and fix direction. Add --json for structured output that plugs into CI and dashboards.
Works with: Terminal, local repos, CI, GitHub
How it fits your workflow
Cheat and Guard sit inside the agent session itself. They plug into the hooks Claude Code already exposes, so Guard can flag a risky action before a tool call runs and Cheat can check a claim right after each step — no rebuild of your workflow required.
The Quality and Audit dashboards work at the repository level. You connect a repository from GitHub, run a scan, and get back a prioritized list of findings with their fixes — on screen for review, and as JSON you can pass straight to your agent to apply. The CLI runs those same two scans from the terminal for local checks and CI.
How to read the findings
Procli produces decision-support, not verdicts. Findings are meant to tell you where to look and what to fix first — they are not a guarantee that software is defect-free, secure, or ready to ship, and human review stays required before you deploy or rely on an output.
Procli is in beta, so Cheat and Guard still surface some false positives, and nothing acts on your code automatically. Every finding leaves the decision with you: dismiss it with a reason, have the agent re-verify or take a safer path, or hold it open until someone addresses it. Audit in particular is scoped deliberately narrow — it targets the common mistakes fast-moving, AI-built apps ship, and does not replace a dedicated security team or a real penetration test.
Open the product
Procli is free during beta while it's shaped with early users; access is currently limited to allowlisted accounts. Once you're in, connect a GitHub repository, then choose Cheat, Guard, Quality, Audit, or the CLI from the dashboard home.
Open dashboard