data-signal-analysis-team

Data Signal Analysis Team Skill

Use this skill when the request is to analyze the workshop dataset, not to build or run an optimization model.

Goal

Produce one autonomous analysis of workshop/data that surfaces:

• data structure and coverage,
• key integrity/consistency checks,
• demand/capacity/cost/risk signals relevant to the disaster-relief planning problem,
• the most important data-driven takeaways.

Do not run a team loop. Do not emit JSON artifacts. Do not generate multiple reports.

Required Tooling

Use MCP server data-explorer-csv:

• describe_csv_schema(file_path) for schema/profile discovery,
• query_csv(file_path, query) for read-only SQL checks.

Rules:

1. Analyze only files under workshop/data.
2. One tool call targets one CSV file.
3. SQL must be read-only and reference csv_data.

Three-Agent Internal Workflow

Agent 1: Scope Agent (what)

• Define the exact analysis objective from the exercise statement.
• Select priority files and checks to answer:
  • Is data complete and coherent for the planning context?
  • What are the strongest operational signals (demand, capacity, arc costs, scenario risk)?
• Create a short checklist of required findings before any conclusion.

Agent 2: Method Agent (how)

• Execute schema discovery on relevant CSVs using describe_csv_schema.
• Run focused query_csv checks to extract evidence.
• At minimum, cover:
  • towns.csv: demand base, critical towns, service targets.
  • depots.csv: capacities, fixed/opening-related signals.
  • arcs.csv: lane availability and shipping cost distribution.
  • scenarios.csv and scenario_demands.csv: uncertainty scale and probability sanity.
  • designs.csv, scenario_flows.csv, scenario_scores.csv: design behavior and risk/cost outcomes.
• Keep results concise and evidence-backed.

Agent 3: Verification Agent (works correctly)

• Verify that all required checks were actually executed.
• Verify no claim is unsupported by a tool result.
• Verify the final report is a single markdown file and contains only high-value findings.
• Remove weak or redundant observations.