/dashboard

Purpose

Phase B of v1.5. Provides a local web dashboard at http://localhost:8501 that visualizes everything the framework already produces on disk. Not a replacement for /status — this is the portfolio-wide view, /status is the current-proposal view.

How it launches

Simple subprocess launch. The dashboard is a Streamlit app at dashboard/app.py.

# From repo root
streamlit run dashboard/app.py

Opens the default browser automatically. Server stops when you Ctrl+C the terminal.

Skill behavior

1. Check prerequisites. Verify Streamlit is importable:

   python -c "import streamlit" 2>/dev/null

   If the import fails, instruct the user to install:

   pip install -r dashboard/requirements.txt

   Then exit.
2. Run the selftest first to catch loader issues before launching:

   python dashboard/selftest.py

   If any warnings appear (e.g., missing evidence ledger, pre-v1.5 proposals), surface them to the user but don't block launch.
3. Launch the app in a subprocess the user can see:

   streamlit run dashboard/app.py

   Report the URL (http://localhost:8501 by default per dashboard/.streamlit/config.toml) so the user can open it manually if the browser doesn't auto-launch.
4. Do not wait for the app to exit. Streamlit is a long-running server. Return control to the user immediately after launching.

Views the dashboard provides

• Portfolio — every proposal in one table with migration status, pipeline progress, compliance coverage, evidence coverage, Gold Team pWin, and spend
• Proposal — per-proposal detail: pipeline, compliance counters, evidence coverage, Gold Team scorecard, recent activity, spend breakdown
• Spend — AI cost aggregated across all proposals, grouped by proposal / skill / model, with top-10 most-expensive runs and daily trend
• Evidence Ledger — browse my-company/evidence-ledger.json, filter by type/status/tags, see citation counts across drafts, surface unused approved evidence

Things to tell the user after launch

• Dashboard reads but never writes. Editing is still done in markdown or via other skills.
• If a proposal shows "pre-v1.5" in the migration column, it just means the JSON sidecars from Phase A aren't present yet — the proposal still works, it just shows less structured detail in the dashboard.
• Token counts are often null in Phase A. Spend shows $0 when counts are null. Real counts come from future SDK integration.

Activity trail

Do NOT append to working/activity.md — this skill is read-only (like /status). If logging the launch is useful for token-spend tracking (it isn't — no AI calls made), that would also be skipped.

When not to use

• During a drafting session inside Claude Code — use /status instead for a quick current-proposal view
• When the user just wants to see one proposal's state
• Before v1.5 Phase A has been run (proposals without JSON sidecars show mostly empty — not useful)