good first issue: announce tool activity to screen readers

[yhay81]

Tool chips in the chat transcript are now human-readable (#19), but they're visual-only: screen-reader users aren't told when the agent starts or finishes a tool step.

Task

• Make new tool-chip activity reach assistive technology — e.g. a polite live region (aria-live="polite") around the chip status, or visually-hidden status text that updates on transitions.
• Announce transitions (a chip appearing, a chip completing), not every re-render — no announcement spam.

Acceptance criteria

• With NVDA or VoiceOver, starting a tool step announces its label (e.g. "Creating a Google Sheet — Q3 budget").
• No duplicate or repeated announcements for the same chip.
• npm test and npm run build stay green.

Reproducible with the bundled sample files — no credentials needed. Please claim with a comment before starting (see CONTRIBUTING.md).

[dubeyharshit0605]

@yhay81 Hi, I’d like to work on this issue. I’ll implement accessible tool-status announcements while preventing duplicate announcements, add or update relevant tests, and run the test and build checks.

[yhay81]

@dubeyharshit0605 — checking in, no pressure attached. You claimed this lane on July 21st; are you still working on it?

Context so you're not blindsided: someone else opened #35 against this issue. I've asked them to hold rather than land over your claim, so the lane is still yours if you want it. If you'd rather hand it off, just say so here and nobody loses time.

If you are still on it, one thing worth knowing before you push: putting the chip's summary text into a live region makes e2e/slack.pw.ts fail intermittently, because getByText then matches both the chip and the announcement. Scoping those assertions to .chat-tool is the cheap fix. Also please route any new user-facing strings through the t/Trans macros — we ship 18 translated locales, and npm run i18n:check won't flag an unwrapped string.