Workflow
- The agent reuses your target, scope, WCAG profile, and authorized actions.
- For websites, it discovers pages and links the inventory with
a1 audit run --inventory. - It observes page or app states and registers ordered user journeys.
- It claims each procedure with
audit next, performs its actions, records current evidence, and evaluates the check. Reader procedures require VoiceOver or NVDA when the procedure specifies a real reader. - It resolves coverage gaps before finalization and builds website reports. Draft reports remain available during the assessment.
a1 audit discover https://createdbyfireside.com --scope site \
--out .a11ied/fireside/inventory.json
a1 audit run https://createdbyfireside.com --environment environment.json \
--inventory .a11ied/fireside/inventory.json --json
a1 audit next .a11ied/fireside/run.json --json
Install the repository's full-site-audit skill through the same
agent skill package as the component development skill.
Resume and hints
Runtime commands own the inventory and assessment state. The agent reads
audit status after interruption and preserves the run's evidence paths.
audit resume blocks interrupted work; explicit retries follow restoration
of a safe UI state. Changed observations revoke old proof. Site hints are hypotheses
to check against the current product, not completed assessments.
Evidence claims
To view an unfinished assessment, use a1 report build --draft with
the run's inventory, page results, and output paths. The draft shows available
results and pending checks without marking the assessment complete. The MCP
report_build tool accepts draft: true.
Unassessed pages stay marked not tested. A partial crawl stays a partial inventory. The report retains recorded procedures, outcomes, evidence notes, and pointers. It reports accepted behavioral judgments with their provenance and artifacts. A passing automated result is not reported as full WCAG conformance.
Virtual observations remain simulated. cantTell, missing capabilities,
blocked journeys, and incomplete discovery remain unresolved. Native app assessment
can use the coordinator, but report publication requires a website inventory.