sandboxed-analysis

name	sandboxed-analysis
description	Generate minimal deterministic Python analyses over typed behavior artifacts, keeping code small, reproducible, and constrained to approved libraries.
metadata	{"annolid":{"always":false}}

Sandboxed Analysis

Use this skill when the bot needs to derive metrics or summaries by generating code.

Rules

1. Generate the smallest analysis program that answers the question.
2. Keep inputs and outputs explicit.
3. Prefer deterministic aggregation over heuristic post-processing.
4. Do not depend on network access or non-approved packages.
5. Preserve replayability: code should be valid against stored artifacts.

Typical outputs

• dwell time,
• bout counts,
• transition matrices,
• proximity summaries,
• assay-specific comparison metrics.