ground

name	ground
description	Full end-to-end pipeline — collect mirror data, run all analysis agents, produce grounding documents.
license	MIT
metadata	{"author":"musicofhel","version":"0.1.0","category":"pipeline"}

When to Use

Use this skill for a complete end-to-end run: collection through grounding document production. This combines the collect and analyze skills into a single orchestrated pipeline.

Full Pipeline

Phase 1: Collection (automated)
  ├── dm-schema         → output/_baseline/stream-schema.json
  ├── dm-chain          → output/*/config/{source,transformed,sent,stored,chain-diff}
  └── dm-collect        → output/*/screenshots, dom, api, timing, meta

Phase 2: Baseline Analysis (1 agent)
  └── baseline agent    → output/_baseline/baseline-report.md

Phase 3: Per-Dashboard Analysis (3 agents × N dashboards)
  ├── Analyst A (Structure) → output/*/analyst-structure.md
  ├── Analyst B (Data)      → output/*/analyst-data.md
  └── Analyst C (UX)        → output/*/analyst-ux.md

Phase 4: Synthesis (1 agent × N dashboards)
  └── Synthesizer           → output/*/grounding.md

Agent Count

For N dashboards: 1 baseline + 3N analysts + N synthesizers = 4N + 1 agents

Example: 6 dashboards = 25 agents total

Prerequisites

• OpenObserve running at OPENOBSERVE_URL
• uv sync && uv run playwright install chromium completed
• Dashboard configs available at DM_CONFIG_DIR

Running

From ~/dashboard-mirror:

# Phase 1 (shell commands)
uv run dm-schema
uv run dm-chain --config-dir ~/dev-loop/config/dashboards
uv run dm-collect

# Phases 2-4 (agent-driven from Claude Code session)
# Baseline first, then per-dashboard analysis, then synthesis

Output

The final deliverable per dashboard is output/<slug>/grounding.md — a comprehensive, fact-based, screenshot-referenced document describing exactly what the dashboard looks like and what state it's in.

Re-running

To refresh a single dashboard:

uv run dm-collect --dashboard <slug>

Then re-run the 3 analysts + synthesizer for that slug only.