Use AgentBay and $architecture-blueprint-generator in this isolated synthetic example. Read .agents/skills/agentbay-skill-installer/SKILL.md and the named provider’s installed SKILL.md and relevant resources before work. For a large skill file, read it in bounded consecutive sections so the instructions are not lost to output truncation. The exact provider source is supplied in source-map.json; this is selected-upstream verification, not a catalogue search. All listed sources were fetched at the recorded commit and Git blob checked. Verify the installed hash against the map. Read brief.md and constitution.md. The fixture fixes scope and waives further discovery questions; never claim a real user interview or human approval. Use only this named provider skill plus AgentBay; do not auto-invoke follow-on skills. Do not delegate, access accounts, install extra dependencies, edit installed skills or checks, run downloaded companion scripts, commit or deploy. Preserve each prior stage’s files.

Review the actual prototype.html and supplied src/storage.js using architecture-blueprint-generator. Write docs/architecture.md (700 words maximum) and docs/architecture.json with components (list of objects with path and responsibility; every path must exist), decisions (at least three objects with choice and tradeoff), and risks (nonempty list of real issues). Include one Mermaid diagram reflecting actual data flow, distinguish the disposable prototype from the planned src/app.js and src/task.js, describe local storage, input validation, async failure/retry and duplicate-submission boundaries. Identify concrete risks and a final-module structure. No invented backend, cloud service, authentication or deployment. Change these two documents only.
