Interaction design · 15 September 2026
Proposed flow · not a completion status
- Step 1User message
- Step 2Discovery updates
- Step 3Interactive choice
- Step 4Server-authorized action
- Step 5Result and next step
On this page
Chat experience and interactive controls
Scope
Use the existing chat experience inside Agent Mate: Markdown rendering, streaming text, interactive choices and correlated activity. The manual project settings panel and chat actions share backend semantics.
Desired experience
When a user shares a repository, show “Discovering…” only while discovery actually runs. Replace generic progress with useful findings as they arrive. Offer project choices and source-access actions as clickable controls; users can also provide details freely in text.
Interactive questions should resolve information the agent cannot safely discover itself. Let users request configuration at any time. If an action is unavailable, identify the specific missing access or configuration rather than responding with a rigid questionnaire.
Decisions
- Use the existing WebSocket channel and established authentication/lease checks.
- Reuse
/chatMarkdown behavior. Treat returned HTML as untrusted content, not executable chat markup. - Bind each control to its project/task and reject stale selections on the server.
- Do not replay mutations on reconnect or imply receipt means completion.
- Display status, errors and uncertainty accurately; accessibility requires text labels alongside color.
- Secret entry is a protected configuration interaction, not a chat message.
Implementation tasks
| ID | Deliverable | Status |
|---|---|---|
| CHAT-01 | Shared rendering and reconnect/error behavior audit | Existing UI; targeted review pending |
| CHAT-02 / MCP-P05 | Project connection/tool selection in panel and chat | Pending |
| CHAT-03 / MCP-P06 | Correlated discovery and tool activity with safe result links | Pending |
Acceptance evidence required
Verify live Markdown rendering, keyboard-operable controls, narrow-screen layout, stale-control rejection, no duplicate mutation on reconnect, and safe rendering of HTML-looking tool output. Existing MCP management WebSocket tests do not establish these Agent Mate interactions.
Related areas
- Agent Mate overview — A guide to the separate designs and how they fit together.
- Architecture and runtime integration — How Agent Mate uses the existing runtime without depending on dashboard lifetime.
- Connections, credentials, and permissions — Shared connection ownership, project grants, key handling and revocation.
- Execution, recovery, and progress tracking — Independent tasks, cancellation, uncertain outcomes and durable progress.
- Project-scoped MCP for Agent Mate — The existing seven-task MCP plan: grants, bridge integration, revocation and acceptance.