| name | analyze |
| description | Run multi-agent analysis pipeline on collected mirror data — baseline, 3 per-dashboard analysts (structure, data, UX), and synthesizer. |
| license | MIT |
| metadata | {"author":"musicofhel","version":"0.1.0","category":"analysis"} |
When to Use
Use this skill after collection is complete and output/ contains mirror bundles. This runs Phases 2-4 of the pipeline.
Pipeline Stages
Phase 2: Baseline Analysis (1 agent)
Runs once across ALL dashboards. Reads:
output/_baseline/stream-schema.json
output/*/config/source.json, sent.json, chain-diff.txt
Prompt template: prompts/baseline.md
Output: output/_baseline/baseline-report.md
Tasks:
- Schema coverage audit (every column in every query vs stream schema)
- Spec compliance check
- Cross-dashboard consistency (naming, colors, queries, granularity)
- Transformation drift (config mutations)
Phase 3: Per-Dashboard Analysis (3 agents per dashboard, parallel)
For each dashboard, spawn 3 independent analysts:
Analyst A — Structure (prompts/analyst-structure.md)
- Layout, sizing, visibility, grid integrity
- Config drift (sent vs stored)
- Output:
output/<slug>/analyst-structure.md
Analyst B — Data & Labels (prompts/analyst-data.md)
- Query correctness, schema coverage, data presence
- Labels, axes, legends, time range behavior
- Output:
output/<slug>/analyst-data.md
Analyst C — UX & Polish (prompts/analyst-ux.md)
- Readability, error states, information hierarchy
- Chart effectiveness, loading performance, color coherence
- Output:
output/<slug>/analyst-ux.md
Phase 4: Synthesis (1 agent per dashboard, after Phase 3)
Reads all 3 analyst reports + baseline + screenshots. Resolves contradictions with visual evidence.
Prompt template: prompts/synthesizer.md
Output: output/<slug>/grounding.md — the canonical grounding document
Invocation Pattern
Each agent receives its full prompt template + specific file paths:
You are Analyst B (Data & Labels). Here is your prompt:
[contents of prompts/analyst-data.md]
Analyze this dashboard:
- Screenshots: output/loop-health/screenshots/
- DOM: output/loop-health/dom/
- API: output/loop-health/api/
- Baseline schema: output/_baseline/stream-schema.json
Write your analysis to: output/loop-health/analyst-data.md
Parallelism
- Baseline must complete before per-dashboard analysts start
- The 3 analysts for each dashboard run in parallel (independent lenses)
- Synthesis runs after all 3 analysts complete for that dashboard
- Multiple dashboards can be analyzed in parallel