CLI reference

-V at the top level prints the package version. --wcag inside a subcommand selects the WCAG version, so the two no longer share one flag name.

sr press takes one chord per argument, and each chord joins its tokens with +: Tab, Shift+Tab, VO+ArrowRight, or NVDA+N.

a1 sr press Tab Tab --sr virtual --json
a1 sr press "VO+ArrowRight" --json

a1 axe http://localhost:3000 --level AA --json
a1 axe http://localhost:3000 --criterion 1.4.3 --json
a1 axe src/components/Dialog.html --rule color-contrast --rule link-name --json
echo '<button></button>' | a1 axe -
a1 axe --html '<img src=x>'
a1 axe http://localhost:3000 --selector '#app' --exclude '.ads'
a1 axe http://localhost:3000 --fail-on critical --json
a1 axe http://localhost:3000 --baseline known-findings.json --update-baseline
a1 axe http://localhost:3000 --format sarif --out results.sarif
a1 axe http://localhost:3000 --format earl --out report.earl.json
a1 axe http://localhost:3000/a http://localhost:3000/b --json

With no selector, the run covers every rule mapped in the WCAG version. The result has violations, passes, incomplete, and inapplicable, plus the resolved rule list under selection.

A file target loads through file://. axe never fetches a URL target's HTML itself, since Playwright loads the page directly.

With more than one target, the JSON result's targets array has one report per target, and text mode prints one report after another.

--format sarif writes a SARIF 2.1.0 log with one rule descriptor per unique axe rule and one result per violating node, merged across every target when more than one is given. It bypasses the envelope entirely, so combine it with --out to feed a static analysis tool such as GitHub code scanning.

--format earl writes an EARL 1.0 report as JSON-LD, the format the W3C reads. The graph is a flat list of assertions, one per failing element, each naming the axe rule that found it and the success criteria that fail with it. Violations become earl:failed, passes earl:passed, incomplete results earl:cantTell, and a run that matched no rule at all becomes a single earl:untested assertion rather than an empty report. Like SARIF it bypasses the envelope, so pair it with --out.

axe exits 4 when a violation remains after --fail-on and --baseline filtering, and 0 otherwise. The JSON envelope's ok field stays true either way. The exit code is the verdict, and the envelope repeats it under result.verdict.

Run with --update-baseline once to accept the current findings, then commit the baseline file so CI only fails on new ones. See understand a violation for what impact means and how to get from a violation to a fix, and script the CLI for the exit codes and the envelope shape.

a1 tree http://localhost:3000
a1 tree http://localhost:3000 --role link
a1 tree http://localhost:3000 --name "Sign in" --json

tree reads Playwright's ariaSnapshot() for the target: the role and name tree a screen reader sees, including JavaScript-rendered content, with no heuristics. Text output is the YAML Playwright produces.

--json parses that YAML into result.nodes, an array of { role, name, level, children } objects.

--role and --name keep a matching node together with every ancestor needed to reach it, so the tree's shape stays intact.

a1 audit http://localhost:3000 --json
a1 audit http://localhost:3000 --fail-on critical
a1 audit http://localhost:3000 --format earl --out report.earl.json
a1 audit --html '<h1>Hi</h1>'

audit runs axe against every rule mapped in the WCAG version, reads the accessibility tree, and derives applicability from the target's HTML, all against one resolved target. The result includes these sections:

result.nextCommands lists a1 wcag rule calls for every failing axe rule.

audit exits 4 when an axe violation remains after --fail-on and --baseline filtering, and 0 otherwise, the same verdict rules as a1 axe.

Checks a11ied cannot automate

Axe checks specific conditions within selected criteria. Assessment procedures cover the remaining behavior and content requirements. An agent records observed results through these commands.

a1 audit pending http://localhost:3000 --level AA
a1 audit record http://localhost:3000 --criterion 2.4.4 --outcome failed \
   --pointer 'nav > a:nth-child(3)' --note "Four links read 'Learn more'."
a1 audit http://localhost:3000 --format earl --out report.earl.json

Pass --results <run-dir>/evidence.jsonl on every audit, pending, and record call in a site audit. Resume with the same file; use a fresh file for a new run. A passing procedure does not erase a failure recorded for another procedure.

--outcome takes passed, failed, cantTell, or inapplicable. Results go to .a11ied/evidence.jsonl, one JSON object per line, so several agent calls can record at once. Recording the same procedure and element again replaces its earlier result. Recording a result never stores it under the axe procedure, so a person's judgment cannot appear in the report as an automated one.

--format earl then writes one report holding both. A recorded result keeps mode: earl:manual or earl:semiAutomatic, so a reader can tell it from an axe result.