Dev Testing CLI for AI Agents

Agent writes code.
Probe verifies it works.

Start servers, check logs, test APIs, control the browser, assert state, record the flow. Replay it forever with zero AI tokens.

$ bun add -g @questpie/probe
Docs
AI agent
_

The feedback loop

AI agent implements a feature. Probe checks it in the running app. If it works — record. Replay on every future change. No manual testing.

01

Code

Agent implements a feature or fixes a bug.

$ git commit -m "add filter"

02

Verify

Probe checks logs, API, browser state.

$ qprobe check && qprobe assert no-errors

03

Record

Capture the working flow as a test.

$ qprobe record stop → filter.spec.ts

04

Replay

Run all tests. Zero AI tokens.

$ qprobe replay --all → 12/12 passed

What it does

PROCESS

Start, stop, restart

Daemon processes with ready detection. PID management. Compose with dependency graph.

LOGS

Aggregate & filter

Timestamped logs from every service. Grep, level filter, follow mode. Merged view.

HTTP

Request & assert

GET/POST/PUT/DELETE with auto baseUrl. Status assertions. JQ filters. Bearer tokens.

BROWSER

Control & inspect

A11y tree snapshots with @refs. Click, fill, type. Console, errors, network.

RECORD

Capture & replay

Record browser actions. Auto-generate Playwright tests. Replay with zero AI tokens.

ASSERT

Verify state

Text, element, URL, title, HTTP status, no-errors, no-network-errors. Exit 1 on fail.

Built with

TypeScript-native CLI powered by the UnJS ecosystem. Minimal dependencies, fast startup, no heavy frameworks.

cittyUnJS
CLI framework
c12UnJS
Config loading
consolaUnJS
Terminal output
ofetchUnJS
HTTP client
tinyexecUnJS
Process spawning
chokidar
File watching
agent-browserVercel
Browser driver
PlaywrightMicrosoft
Test replay
tsdownRolldown
Build tool
Biome
Lint & format

Close the feedback loop.

Install the CLI. Add the skill to your agent. The next feature your agent writes gets verified automatically — logs checked, API tested, browser inspected, test recorded.

Read the DocsGitHub