Command Risk Assessment

Every terminal command is scored with a 4-level risk system before execution.

Risk Levels

Level	Auto-Approved?	Examples
Safe	Yes (if prefix whitelisted)	ls, cat, date, echo
Medium	After first approval	curl, git push, npm install
High	After first approval	rm, mv, chmod, kill
Critical	Never persisted (default)	sudo rm -rf, dd, mkfs

Permission Modes

Key	Type	Default	Description
default	—	—	Safe/Medium/High approvals persist across restarts. Critical is per-session only.
cautious	—	—	All approvals are per-session only.
yolo	—	—	All approvals persist forever.

Dangerous Constructs

Patterns that always elevate risk:

• Command substitution: $(...) and backticks
• Process substitution: >(...), <(...)
• Redirection: >, >>
• Multiple commands: ;, &&, ||
• Pipes into shells or sudo

Sensitive Path Detection

Commands referencing these files are automatically elevated:

• .env — environment secrets
• SSH keys: id_rsa, id_ed25519
• Cloud configs: .aws, .kube, .docker
• System auth: shadow, passwd, sudoers
• Credentials: master.key, .netrc, .pgpass

Hard Blocks (New)

Even before approval flow, broad or sensitive delete patterns are blocked, including risky rm -rf and find ... -delete targets.