Environment

Agent environment

Coding agents inherit the developer's shell environment. That can include API keys, cloud credentials, local config paths, SSH agent access, package manager settings, and proxy variables.

Rye supervision is most reliable when the agent process is launched through Rye, because the process sees proxy and trust settings at startup.

Recommended launch pattern

Use:

rye wrap claude

rye wrap codex

For repeated use, install shims:

rye app wrap claude

rye app wrap codex

Keep scope explicit

Avoid global proxy settings when you are testing. A targeted wrapper makes it easier to reason about what Rye is supervising and which process produced each audit record.

Files and secrets

Agents can read local files when the runtime permits it. Policy should account for files such as:

• .env
• cloud credential files
• SSH config
• package manager tokens
• Terraform state
• customer exports

The goal is not only to prevent obvious key leakage. It is to reduce accidental movement of sensitive context into model requests.