CLI commands

Every verifaied command and its main options. Run any command with --help for the full list.

Install

uv tool install verifaied

Or, with pipx:

pipx install verifaied

Then log in once with verifaied login. Set up walks through the rest.

Environment variables

VariableWhat it does
VERIFAIED_API_TOKENAn API token. Wins over a stored login. This is how CI authenticates.
VERIFAIED_API_URLThe API to talk to, when it is not the hosted one. Each API URL keeps its own login.

Create a token under Settings → API tokens. See API tokens.

Options most commands take

Commands that talk to verifAIed work out the repository and branch from the directory you run them in. These options override that:

OptionWhat it does
--repo, -rThe repository, as owner/name. Default: read from the git remote named origin.
--branch, -bThe branch. Default: the branch checked out.
--api-urlThe verifAIed API to talk to.
--tokenAn API token to use instead of the stored login.

Account and setup

CommandWhat it doesMain options
verifaied loginLog the CLI in. Opens an approval page in your browser and stores the login on this machine.
--check say whether this machine is logged in, and log nothing in
--no-browser print the link instead of opening it
verifaied logoutRevoke this machine's login and forget it.
verifaied setup <agent>Connect a coding agent. claude is set up for you: both MCP servers and the skill. codex, cursor, antigravity and other print the steps to do by hand.
--scope for Claude Code: user (the default), local or project
--force replace what is already there
verifaied skill installInstall the verifAIed skill into this repository's .claude/skills, so your agent knows the workflow.
--user install for every repository on this machine
--dir install into a skills directory you name
--force overwrite an installed copy that differs

Coverage

CommandWhat it doesMain options
verifaied uploadUpload a coverage report and the source files it names for the current branch. Reads pytest-cov coverage.json and Istanbul coverage-final.json.
--coverage a report to send; repeat it to send several
--junit a JUnit XML file, for failing-test detail
--browser upload as code reached instead of unit coverage
--root the directory the report's paths are relative to
--dry-run print what would be sent, and send nothing
verifaied checkList untested and partially covered functions from a local coverage.json. Python only. Runs on your machine and needs no account.
--coverage the report to read (default ./coverage.json)
--prompt print a test prompt for each function
--limit how many functions to list; 0 for all
--fail-under exit 3 below this percentage of fully covered functions
verifaied check-doneAsk whether the current branch is done, and list the gaps that are left.
--json print the raw verdict
--allow-warnings exit 0 when only non-blocking gaps are left

Browser tests

CommandWhat it doesMain options
verifaied test studioConnect this machine to verifAIed and leave it running, so you can record and run tests from the web.
--env the environment to run against
--branch pin the branch instead of following the one checked out
verifaied test record "<folder>/<name>"Record one test from the terminal: opens a browser for you to click the flow through.
--url where the flow starts
--pre another test to replay first; repeatable
--description what the flow is for
--env the environment to run against
verifaied test edit "<folder>/<name>"Keep the first steps of a test and record the rest again.
--from the step to start again from; 0 redoes the whole test
--env the environment to run against
verifaied test run "<folder>/<name>"Replay a test and save its video and result. Name one test, or use --folder or --all.
--folder run every test under one folder
--all run every recorded test
--retries replay a failed test up to this many more times
--parallel replay up to this many tests at once
--video how the video is paced: fast, watch or off
--env the environment to run against
--proxy measure code reached in an app with no build step
verifaied test listList the recorded tests and how each last went on this branch.
verifaied test syncCreate or update a test in verifAIed for every spec under .verifaied/tests. Never deletes one.
--dry-run say what would change, and send nothing
verifaied test mcpServe the recorder to a coding agent as a local MCP server. Your agent starts it; you do not run it yourself.
--env the environment to record against
verifaied var set <name> <value>Set a variable for an environment in .verifaied/config.toml.
--from-env read the value from an environment variable; use this for secrets
--unique add a fresh suffix to the value on every run
--env the environment to set it in
verifaied var unset <name>Remove a variable from an environment in .verifaied/config.toml.
--env the environment to remove it from

Code reached and visual checks

CommandWhat it doesMain options
verifaied instrumentPrint the setup that lets a browser measure your app, for the stack it finds. Edits nothing.
--check print nothing; exit 0 if instrumented, 3 if not, 4 if it cannot tell
--ecosystem choose the stack: vite, babel or static
--root the app directory to inspect
verifaied collectMeasure the code a browser session reaches. Alone, it opens a browser window for you to click through. With a command after --, it wraps that command.
-- <command> the suite to wrap, for example -- npx playwright test
--url where the browser window opens
--proxy instrument an app with no build step through a local proxy
--no-browser listen only, and open no window
--test-results where the wrapped suite writes traces and screenshots
verifaied clear-manualReset what manual sessions reached on a branch. Every function goes back to Unexercised.
verifaied audit <url>Load a page at several widths and report overflow, low contrast, broken images and console errors. Needs the verifaied[audit] extra.
--viewports comma-separated widths (default 375,768,1280)
--driver who is running it: human, agent or suite

Exit codes

Most commands follow one convention, so you can use them as a gate in CI:

  • 0: success.
  • 1: usage or configuration error, such as a missing file or no login.
  • 2: the API could not be reached or refused the request.
  • 3: the check failed: gaps remain, findings were recorded, or coverage is too low.
CommandExit codes
verifaied check-done0 done. 3 blocking gaps remain. 2 API error. 1 usage error.
verifaied test run0 every test passed. Otherwise the worst exit code of the tests it ran. 2 a run passed but its upload failed. 1 usage or configuration error.
verifaied audit0 clean. 3 findings recorded. 2 the audit could not be run or recorded. 1 usage error.
verifaied collectWith a wrapped command, that command's own exit code when it failed.