refactor(ai-elements): wrap fast agent composer

[HomenShum]

What changed

• Wraps the FastAgentPanel composer with AI Elements PromptInput primitives while preserving NodeBench send, stop, prompt-enhancement, and product-intake behavior.
• Adds a deterministic parent send contract so attachments and documents are either delivered to a capable backend or held fail-closed.
• Keeps analyzing documents attached and unsent, passes real ready/legacy document IDs, and uploads real product files before beginRun.
• Restores keyboard recovery: the panel skip link targets the actual textarea, and attachment removal is visible and operable on focus.
• Adds parity, send-contract, Enter-key, document, attachment, product-mode, and accessibility regressions.

Why

The previous composer could not adopt the shared primitive safely without an explicit boundary between text-only and attachment/document-capable backends. Independent review caught two additional P1 accessibility defects: a non-focusable skip-link target and a mouse-hover-only attachment recovery control. Both were reproduced failing-first and are fixed.

Validation

• Focused InputBar/send contracts: 26 passed
• Full FastAgentPanel: 173 passed, 19 environment-gated skipped
• Explicit streaming guard: 1 passed
• npx tsc --noEmit --pretty false
• Design system: 11 passed; npm run lint:design has 0 high findings
• npm run build
• Bundle guard: FastAgentPanel 1,550,259 B; separate code block 73,341 B; separate Streamdown 163,187 B; 0 Shiki/oniguruma/wasm markers in panel; 0 assets/shiki/ service-worker precache entries
• QA memory: f904f7c63d19 and a7aaecb7821b fixed, 0 open/regressed

Release

Ready for required checks and CI-gated squash auto-merge. No admin bypass.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
nodebench-ai	Ready	Preview, Comment	Jul 14, 2026 12:19pm

[github-actions]

PR size advisory

This PR adds 1343 lines of substantive change. CONTRIBUTING.md defines a soft limit of ~400 LOC.

If the PR is genuinely cohesive (e.g. an architecture map, a generated migration, a deletion of a dead module), no action is needed. Otherwise consider:

• splitting into 2-3 PRs along independent concerns
• pre-discussing the architecture change in a GitHub Discussion before merge

This is advisory — it does not block the merge.

[github-actions]

✅ Dogfood Visual QA Gate: PASSED

Check	Status
Screenshots	23 captured (pass)
Walkthrough	9 chapters (pass)
Key Frames	9 extracted (pass)
Scribe Steps	8 how-to steps (pass)
Build	success

Artifacts

Download the dogfood-evidence-9f29032 artifact from the Actions tab for full screenshots, frames, and walkthrough video.

---

Generated by Dogfood QA Gate

[github-actions]

Demo: walkthrough of the surfaces this PR changed is available as a workflow artifact (pr-demo-526) at https://github.com/HomenShum/NodeBenchAI/actions/runs/29332368401