feat: Agent Output Canvas — rich HTML documents via Mini App

Problem

Long agent outputs (research reports, analyses, comparisons) are delivered as multiple truncated Telegram messages. Telegram's 4096-char limit, broken tables, and lack of navigation make complex documents unusable in chat.

Proposed Solution

Agent Output Canvas — when an agent produces a large or richly-formatted document, it generates an HTML page served by Right Agent and sends a single button to the user that opens it as a Telegram Mini App.

User Flow

User: "Research AI agent trends and write a full report"

Barry: [works for 3 minutes]

Barry: 📄 Research complete → [Open Report]  ← one Telegram message

[User taps] → Mini App opens with full HTML report:
  - Table of contents
  - Sections with headings
  - Formatted tables
  - Syntax-highlighted code
  - Collapsible sections

Features

Core

• Agent writes HTML/Markdown to a unique URL on the Right Agent server
• Mini App renders it with proper styling (Telegram theme-aware)
• Single button message in chat instead of flooding

Rich Formatting

• Tables, nested lists, code blocks with syntax highlighting
• Collapsible sections for long content
• Auto-generated Table of Contents with anchor links
• Responsive mobile layout

Output Templates

• Research Report
• Meeting Notes
• Code Review
• Comparison Table
• Task Breakdown

Live Updates

• Agent streams content to Canvas as it writes (SSE/WebSocket)
• User sees document being built in real time
• Progress indicator while agent is still working

Sharing & Export

• Public shareable URL (no auth required for read-only access)
• Export to PDF / copy as Markdown
• Optional: user can edit the canvas, agent sees changes

Why This Is Unique

All major agent frameworks (Hermes, Claude Code, AutoGen) dump output into a terminal or file system. There is no concept of a "living document" tied to the agent's output that:

• Opens inside the messaging app the user is already in
• Updates in real time as the agent writes
• Is shareable without any login

Technical Notes

• Right Agent HTTP server serves canvas pages at /canvas/{id}
• Agent writes to canvas via new MCP tool: mcp__right__create_canvas(title, content_html) + mcp__right__append_canvas(id, html)
• Mini App auth via Telegram initData for private canvases
• Public canvases use a signed token in URL (no auth needed)
• Canvas storage: SQLite or flat files in agent's working directory

🤖 Generated with Claude Code

Start with the Right Agent HTTP server and the proposed /canvas/{id} route, then trace how MCP tools could create and append canvas content. Review the Telegram Mini App auth and storage options described in the issue; done means a rendered, navigable canvas can be opened from one chat button and updated by the agent.