result-analyst

name	result-analyst
description	Spawn an isolated result-analyst subagent to read docs/ and outputs/ and produce a structured experiment summary with metric tables, key findings, and next-step recommendations.
tags	["m2g-net","results","analysis","subagent"]
version	1.0.0
user-invocable	true
allowed-tools	["Agent"]

Result Analyst — Subagent Launcher

Spawns an isolated read-only subagent that analyzes M2G-NET experiment outputs. No files are modified. The subagent only reads and summarizes.

When to Invoke

• After completing a training + evaluation run
• When you want a summary of what the latest experiments show
• Before writing the results section of the manuscript
• When comparing multiple experiment variants

How to Invoke

Agent(
  description="Analyze M2G-NET experiment results",
  subagent_type="general-purpose",
  prompt=[contents of .agents/subagents/result-analyst/agent.yaml default_prompt]
)

What it Analyzes

• All .md files in docs/ matching *RESULTS*.md, *REPORT*.md, *SUMMARY*.md
• All .csv files in outputs/
• Identifies best model variant, regressions, and anomalies

Output

Returns a structured report under 400 words:

• Metric table (ROC-AUC, ECE, F1 per model variant)
• Key findings with file references
• Regressions vs. prior runs
• Recommended next experiments (specific, actionable)