Team Mode: capture design ideas for privacy-safe team coaching signals

Thank you @vanditmakwana55 for the thoughtful work in #56! The idea of giving teams visibility into aggregate coaching signals — while keeping raw prompts, code, and file names strictly local — is a compelling direction worth exploring further.

This issue captures the design ideas surfaced in that PR so we can reference them in future planning.

Design ideas worth preserving

1. Privacy-safe snapshot model — only aggregate numeric stats leave the developer's machine: practice-group scores, token totals, anti-pattern counts by severity, and week-over-week deltas. No raw prompts, source code, file names, or session transcripts.

2. Opt-in by default — Team Mode is off unless the user explicitly enables it and provides a server URL + developer ID.

3. Local snapshot assembler — converts existing analyzer outputs into a compact numeric payload after each analysis run, keeping the sync layer decoupled from the analysis pipeline.

4. One-way sync client with retry — snapshots are posted asynchronously and non-blocking; failed uploads are retried quietly without affecting the local experience.

5. Team Dashboard UI — a webview page showing team-wide scores, trends, and per-member rows with filtering by category and date range. Includes PDF export.

6. Practice-group scoring — five categories (prompt quality, session hygiene, code review, tool mastery, context management) each scored 0–100.

Current project direction

We do not plan to introduce a server-side component (backend API, hosted database, or cloud infrastructure) as part of this project. The AI Engineering Coach is designed to run entirely within the VS Code extension, and we want to keep it that way for simplicity, trust, and ease of adoption.

That said, the client-side aspects of this design (local snapshot assembly, privacy boundaries, dashboard visualization) contain ideas we may draw from in the future — for example, enabling local-only team reports or file-based sharing workflows that don't require a running server.

Next steps

• Keep this issue open as a reference for future team-oriented features
• Any future work in this space should preserve the privacy-first, no-server constraint

This issue names no implementation files, tests, or entry points; start by reviewing PR #56 and the existing analyzer outputs. The documented direction is only to preserve privacy-safe, client-side ideas without a server, so a concrete implementation scope and definition of done would need to be established first.