reticlehq/reticle

What it solves

Reticle provides a verification layer for AI coding agents to prevent them from shipping broken code that "looks" correct. It solves the problem of "silent failures"—such as 500 errors, incorrect state values, or console errors—that are hidden from the UI and therefore invisible to agents relying on screenshots or DOM-only tools.

How it works

Reticle runs a dev-only SDK inside the running web application. Instead of observing the app from the outside, it monitors internal program events including network calls, store state, signals, and the React fiber tree. When an agent performs an action, Reticle checks these internal events against a set of predicates (e.g., "did the POST request return a 200?") and returns a deterministic pass/fail verdict along with the specific file:line in the code that needs fixing.

Who it’s for

Developers using AI coding agents (such as Claude Code, Cursor, Copilot, or any MCP-compatible agent) to build web and desktop applications using frameworks like React, Vue, Svelte, Preact, Astro, or Electron and Tauri.

Highlights

• Internal Visibility: Accesses network responses, app state, and console logs that are invisible to vision models and browser agents.
• Deterministic Verification: Uses structured events rather than LLM guessing or screenshots, eliminating flake.
• Agent-Integrated: Works as an MCP server and provides a slash command (/reticle) for seamless integration into the agent's loop.
• Direct Attribution: Provides the exact file:line for failures, reducing the number of tool calls needed for an agent to fix a bug.
• Regression Testing: Allows recording a flow once and replaying it deterministically on every subsequent edit.