[Aaryan430]

Stacked on #566: merge that first, then this PR's base moves to main.

Why

• The score was dropped. Asked for a scan's accessibility score, or for a plan to raise it, the agent said the report has no score. The scan run summary does return one (scoreData). The MCP already reads that summary, but kept only the run settings and the changes since the last run.
• Plans came from the agent, not BrowserStack AI. With nothing pointing it to askBrowserStackAI, the agent made up its own plan from one page of issues. BrowserStack AI answers from the whole report and the product's own knowledge, so report questions should go there.

What changed

The score is given with the issues

• On the first page of fetchAccessibilityIssues and on a finished startAccessibilityScan (a new scan or a re-run), a line gives:
  • the whole report's score, as the dashboard shows it;
  • the share of points each severity lost, and the passed/failed checks per severity;
  • how the score is worked out, from the backend's scoring code:
    • a check is one rule on one page, and best practices are not scored;
    • a check fails when any element on the page fails the rule;
    • points per check by severity: critical 10, serious 7, moderate 3, minor 1;
    • a failed check counts double, and the score is the passed share of all points.
• Read from the report's own summary (GET …/scan_runs/{id}, …/builds/{id}, …/reports/{id}) for Website Scanner runs, Automated Tests builds and Workflow Analyzer reports.
• A scanner run read by its build id has no scan ids, so it is found in the scan list first. That list knows only a scan's latest run, so an earlier run read by its link has no score.
• No line when the report has no score: the plan has none, the report is not finished, the summary cannot be read, or it is an app build or report.
• Later pages don't read the score again.

Report questions go to BrowserStack AI

• The same first-page results end with a line telling the agent to call askBrowserStackAI with product: a11y, the user's question and the report, for anything beyond listing its issues (a plan, what to fix first, why it scored as it did), and not to work out the plan itself.
• If BrowserStack AI says it is not enabled for the account, the agent tells the user and answers from the issues and score.
• The line is in the tool's result, not a description, so no description budget changes.

Verified

• tsc, eslint and prettier (src) pass; vitest passes (151 files, 2269 tests), on top of fix(accessibility): land #541 and #553 on main #566.
  • New tests: the score for each report type, a scanner run found by its build id, no score (missing, unreadable, app reports), the score and BrowserStack AI lines on the first page only, and on startAccessibilityScan.
• Live, in isolated claude -p sessions against a real scan scoring 52, two runs of each question:

Question	Before	After
"What's the accessibility score of my scan?"	"I couldn't get an accessibility score"	"52 out of 100", with the per-severity breakdown
"How can I improve the accessibility score for my scan? Make me a plan."	The agent's own plan; "the report doesn't give a score number"	Calls askBrowserStackAI with the question and the report

• Not verified end to end: BrowserStack AI's answer through the MCP. Every askBrowserStackAI call currently fails with a CDN 503 first byte timeout after about 60 s, on this branch and on the released 2.6.1 alike, while the dashboard's Ask AI answers the same question for the same account. That is being raised separately. Until it is fixed, the agent tells the user BrowserStack AI failed and falls back to its own plan from the issues and score.

🤖 Generated with Claude Code

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Central YAML (base), Organization UI (inherited), Workspace UI (inherited); Enforced overrides: Organization
• Review profile: ASSERTIVE
• Plan: Enterprise
• Run ID: 3b8dc23b-a1aa-4998-8377-d1d65ee006a6

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts