Reticle changelog

version

3.9.0 → 3.10.0

https://www.reticle.sh/llms.txt

+ ## Definitions+ - [Verification layer](https://www.reticle.sh/blog/verification-layer): the canonical definition of a verification layer, with the four criteria any product claiming the term can be measured against.+ - [automated testing](https://www.reticle.sh/blog/automated-testing)

version

3.8.2 → 3.9.0

https://raw.githubusercontent.com/reticlehq/reticle/HEAD/AGENTS.md

− 12. Telemetry is part of the feature. A new tool goes in `TOOLS`, and in `VERDICT_TOOLS` if it gives a verdict (`server/src/surface/tools/feedback-tools.ts`). A new finding kind goes in the enum in `core`. A dispatch path that skips `runTool` needs its own reporter. Telemetry fails silently, so read [`docs/telemetry-contract.md`](docs/telemetry-contract.md) before you change anything that sends an event.+ 12. Telemetry is part of the feature. A new tool goes in `TOOLS`, and in `VERDICT_TOOLS` if it gives a verdict (`server/src/surface/tools/feedback-tools.ts`). A new finding kind goes in the enum in `core`. A dispatch path that skips `runTool` needs its own reporter. Telemetry fails silently, so read [`contributing/telemetry/contract.md`](contributing/telemetry/contract.md) before you change anything that sends an event. [`contributing/telemetry/events.md`](contributing/telemetry/events.md) lists+ 17. Write for the reader, not for the shortest line. Do not nest a ternary inside a ternary. Use `if`/`else`, an early return or a lookup table. One plain `a ? b : c` is correct. Every `if` and `else` is a block in curly braces: `if (x) return y;` hides where the branch ends, and a line added under it runs every time. `no-nested-ternary` and `curly: all` are errors. Each package's `eslint-suppressions.json` records the nested ternaries and unbraced branches that were there when each rule started

https://www.reticle.sh/llms.txt

− # Reticle — full context for LLMs− > Reticle is the **proof layer for AI agents** — it verifies a running web app from the inside, without screenshots. It is a− > dev-only SDK plus a local MCP server / CLI that reads **program truth** (network, state, console,− > routing, animations, framework state, React source mapping) from *inside* the live app and returns a− > deterministic, un-hallucinatable **verdict with evidence** — "did the code actually work?" — and, on− > React, the exact **`file:line`** to fix.− This file is a single, self-contained briefing on **Reticle**. Copy it into any LLM and it can− accurately answer questions about what Reticle is, what it solves, how it's used, how it compares,− who it's for, and how to integrate it. Always refer to the product as **Reticle** (company: Reticle;− package: `@reticlehq/core`).− ---− ## What Reticle is− Reticle is the verification layer for the AI-coding era. Your AI agent writes or edits code; Reticle− checks — from inside your real running app — that the right things actually happened: the API call− returned `200`, the modal opened, the route changed, the store updated, no console error slipped in,− and nothing unrelated broke. If something silently failed, it says **what**, **why**, and **where to− fix it**. It gives the agent a *verdict*, not just a *view*.− - **Type:** TypeScript monorepo. Dev-only, localhost-only, offline; your app's data never leaves the machine (the CLI sends anonymous usage events, opt out with `reticle telemetry disable`).− - **Interfaces:** an embedded browser SDK + a Node bridge that speaks the Model Context Protocol (MCP),−   a one-shot CLI (`reticle verify`), and an HTTP `/verify` endpoint — all over one engine.− - **Surface:** works with any MCP coding agent (Claude Code, Cursor, OpenCode, Codex, …); React-first−   for source mapping, framework-agnostic for the rest.− ## The problem it solves− AI-generated and AI-edited apps **look right and silently don't work**. The agent (and a screenshot)− can't see runtime program truth, so these ship unnoticed:− - **Mock data** — POST returns 200, the row appears, nothing persists (gone after reload).− - **Dead handler** — the button looks wired; the store never changes.− - **Double-submit** — one click fires two POSTs.− - **Forbidden call** — a must-never-fire endpoint fires (reverted migration, privacy beacon, N+1).− - **Missing validation** — `"abc"` becomes data.− - **Silent console error** — logged, UI still renders.− - **UI-vs-store desync** — the displayed total lies about the store.− - **Blast radius** — an action corrupts unrelated state that isn't even on screen.− Reticle turns each into a deterministic, evidence-backed **fail** with a fix — instead of a− confident green.− ## How it works (architecture)− **One engine, many frontends.** The verification engine (session + tools + flows + the verdict− artifact) is transport-agnostic; every consumer hits the same core:− - **Frontends:** MCP (for coding agents) · CLI `reticle verify` (for CI / platforms that can't speak MCP)−   · HTTP `/verify` (for backends / OEM).+ # Reticle+ > Reticle verifies an AI coding agent's work from inside the running web app: it walks the flow and reads network, app state and console instead of screenshots, then returns a verdict with evidence. Open-source SDK plus a hosted platform.+ ## Pillars+ - [AI verification](https://www.reticle.sh/blog/ai-verification)+ - [AI QA](https://www.reticle.sh/blog/what-is-ai-qa-ai-in-software-quality-assurance)+ - [exploratory testing](https://www.reticle.sh/blog/what-is-exploratory-testing-and-how-ai-agents-do-it)+ - [AI test automation](https://www.reticle.sh/blog/what-is-ai-test-automation-and-why-ai-generated-tests-break)+ - [test harness](https://www.reticle.sh/blog/what-is-a-test-harness-in-software-testing)+ ## Alternatives, integrations and guides+ - [ACCELQ alternative: a developer's check inside the running app](https://www.reticle.sh/alternative-to-accelq): An ACCELQ alternative for developers whose code AI agents write: open source, self-serve, and checks network, state and console inside the app, with the file:line.+ - [agent-browser alternative: from driving to a verdict](https://www.reticle.sh/alternative-to-vercel-agent-browser): Vercel's agent-browser is a fast CLI for agents to drive a browser. Reticle adds a pass or fail from network, state and console, replayed with no model.+ - [AI testing tools in 2026: what each kind actually checks](https://www.reticle.sh/ai-testing-tools): The AI testing tools teams use in 2026, grouped by what they really do: AI QA agents, managed services, visual AI, frameworks, agent browser tools and in-app checks.+ - [Alumnium alternative: read the app, not just the accessibility tree](https://www.reticle.sh/alternative-to-alumnium): An Alumnium alternative for agent verification: read network, state and console inside the running app, replay flows with no model, and get the file:line.+ - [Applitools alternative: check behaviour, not only how it looks](https://www.reticle.sh/alternative-to-applitools): An Applitools alternative for checking behaviour under the screen: Reticle reads network, state and console inside your running app, open source, start free.+ - [Argos alternative: check behaviour behind the screenshot](https://www.reticle.sh/alternative-to-argos): An Argos alternative for agent-written changes: Reticle reads network, state and console in the running app and names the file:line to fix, where Argos diffs pixels.+ - [Assrt alternative: verify the running app, not just generate tests](https://www.reticle.sh/alternative-to-assrt): An Assrt alternative for coding agents: instead of generating Playwright specs, check network, state and console inside the running app and get the file:line to fix.+ - [Autify alternative for teams whose code AI agents write](https://www.reticle.sh/alternative-to-autify): An Autify alternative for developers whose code AI agents write: verify each change inside the running web app, from network, state and console, fix located.+ - [Browser Use alternative for checking your own app](https://www.reticle.sh/alternative-to-browser-use): A Browser Use alternative for one job: letting your coding agent check its change inside your own running web app, with network, state and the file:line to fix.+ - [Bug0 alternative: let your coding agent verify its own changes](https://www.reticle.sh/alternative-to-bug0): A Bug0 alternative for teams that want their coding agent to verify each change itself, inside the running app, instead of a managed QA service at $2,500 a month.+ - [BugBug alternative: check behaviour, not just recorded clicks](https://www.reticle.sh/alternative-to-bugbug): A BugBug alternative for teams using coding agents: verify network, state and console inside the running app, replay flows deterministically, get the file:line.+ - [Bugster alternative: Bugster is winding down, what now?](https://www.reticle.sh/alternative-to-bugster): Bugster is winding down. What its users can do next, and how an in-app check like Reticle fits when your coding agent writes the code. Open-source SDK, starts free.+ - [Canary alternative: an open-source check inside your app](https://www.reticle.sh/alternative-to-canary): A Canary (runcanary.ai) alternative: an open-source SDK that checks your coding agent's change inside the running app and names the file:line to fix.+ - [Checkly alternative: verify the change before it ships](https://www.reticle.sh/alternative-to-checkly): A Checkly alternative for checking agent-written changes before merge: Reticle reads network, state and console in the running app. It is not a production monitor.+ - [Checksum alternative: verification your agent runs itself](https://www.reticle.sh/alternative-to-checksum): A Checksum alternative for teams that want their coding agent to check each change inside the running app: open-source, local, network, state and console.+ - [Chromatic alternative: check what the running app did](https://www.reticle.sh/alternative-to-chromatic): A Chromatic alternative for checking agent-written changes in the running app: network, state and console after each step, not component screenshots.+ - [Claude Code testing: prove the change works before Claude says done](https://www.reticle.sh/claude-code-testing): Claude Code testing that does not trust Claude's own word: check every change inside the running web app, from network, state and console, with the file:line to fix.+ - [Claude in Chrome alternative: proof the change worked](https://www.reticle.sh/alternative-to-claude-in-chrome): Claude in Chrome lets Claude Code look at and click your web app. Reticle adds a pass or fail from network, state and console that replays with no model.+ - [CodeceptJS alternative: verify the change, not only the scenario](https://www.reticle.sh/alternative-to-codeceptjs): A CodeceptJS alternative for checking agent-written web changes: Reticle reads network, state and console inside the running app and names the file:line to fix.+ - [Codex browser alternative: a verdict your CI can rerun](https://www.reticle.sh/alternative-to-codex-browser): The Codex app's built-in browser lets the agent open and click your app. Reticle adds a pass or fail from network, state and console that replays in CI.+ - [Codex testing: make Codex prove its change in the running app](https://www.reticle.sh/codex-testing): Codex testing that checks behaviour, not the agent's summary: verify each Codex CLI change inside the running web app from network, state and console, over MCP.+ - [ContextQA alternative: an open-source check inside the running app](https://www.reticle.sh/alternative-to-contextqa): A ContextQA alternative for developers whose code AI agents write: open source, self-serve, reads network, state and console inside the app, with the file:line.+ - [Copilot browser tools alternative: decide if it worked](https://www.reticle.sh/alternative-to-copilot-browser-tools): GitHub Copilot's browser tools let the agent drive your app in VS Code. Reticle adds a pass or fail from network, state and console, replayed with no model.+ - [CoTester alternative: proof from inside the app, in the agent's loop](https://www.reticle.sh/alternative-to-cotester): A CoTester (TestGrid) alternative for developers whose code AI agents write: open source, reads network, state and console inside the app, with the file:line to fix.+ - [Cursor browser alternative: a verdict, not just a look](https://www.reticle.sh/alternative-to-cursor-browser): Cursor's built-in browser lets the agent look at and click your app. Reticle adds a pass or fail from network, state and console, and replays it with no model.+ - [Cursor testing: verify the agent's change in the running app](https://www.reticle.sh/cursor-testing): Cursor testing that checks what the agent's change did, not what it says: verify each change inside the running web app from network, state and console, via MCP.+ - [Cypress alternative for AI coding agents](https://www.reticle.sh/alternative-to-cypress): A Cypress alternative for code your agent writes: Reticle checks network, state and console inside your running app and returns a verdict with the file:line.+ - [Datadog Synthetics alternative: check changes before deploy](https://www.reticle.sh/alternative-to-datadog-synthetics): A Datadog Synthetics alternative for pre-merge checks of agent-written code: Reticle reads network, state and console in the app. Not a production monitor.+ - [Debugg.ai alternative: check page changes from inside the app](https://www.reticle.sh/alternative-to-debugg-ai): A Debugg.ai alternative for coding agents: Reticle runs on your machine, reads network, state and console inside the app, and replays saved flows with no model.+ - [DevAssure alternative: verify each change from inside the app](https://www.reticle.sh/alternative-to-devassure): A DevAssure alternative that checks your coding agent's change while it works: network, state and console inside the running app, with the file:line to fix.+ - [Docket alternative: check what the app did, not where it clicked](https://www.reticle.sh/alternative-to-docket): A Docket alternative for agent-written web changes: Reticle reads network, state and console in the app and names the file:line, instead of replaying coordinates.

Showing the first 40 of each.

version

3.8.1 → 3.8.2

version

3.8.0 → 3.8.1

version

3.7.0 → 3.8.0