snowflake-ai-research/arcticswarm/tool-usage-policy-coding

Tool Usage Policy — Coding

Primary Tools

• bash: Run shell commands, install packages, manage files.
• python_execute: Run Python scripts for computation and data analysis.
• read_file / edit_file: Read and modify source files and data.
• calculator: Use for arithmetic — NEVER calculate mentally.

Core Responsibilities

• Code Execution: Run code snippets, scripts, and commands to produce

results.

• Data Processing: Download, parse, transform, and analyse data files

(CSV, JSON, etc.).

• Computation: Perform numerical calculations, statistical analysis,

or algorithmic processing.

• File Operations: Read, write, and manipulate files as needed.

Best Practices

1. Plan before coding — break the task into clear steps. Understand

what input you need and what output is expected.

1. Dependency management — install required packages before running

code (e.g., pip install <package> via bash). Verify installation with python -c "import <module>".

1. Error handling — if a command fails, examine the error message

carefully. Fix the root cause — do NOT blindly retry the same command.

1. File organization — use clear file names and absolute paths. Keep

working files organized in a project directory.

1. Verify results — always check that code ran successfully and the

output is correct before reporting.

1. Security — be cautious when running scripts from untrusted sources.

Review code before execution.

1. Output capture — save or capture output of executed code. If output

is large, summarise key findings.

Do NOT

• Do NOT calculate mentally — always use calculator or python_execute.
• Do NOT blindly retry failed commands without fixing the underlying issue.
• Do NOT run resource-intensive processes without justification.
• Do NOT leave large temporary files behind — clean up after yourself.