shell

func ClassifyEffect ¶ added in v0.6.3

func ClassifyEffect(args map[string]any) tool.Effect

ClassifyEffect maps a shell-tool invocation to one of three effect tiers: - EffectDangerous → prompts the user before running - EffectReadOnly → runs in plan mode (no prompt) - EffectMutates → runs in normal mode without prompt, blocked in plan mode

func IsDangerousCommand ¶ added in v0.6.3

func IsDangerousCommand(command string) bool

IsDangerousCommand reports whether the command requires user confirmation before running. Kept narrow on purpose: only commands that are hard to reverse or that escalate privileges. Routine mutations (chmod, kill, find -delete, ...) are not dangerous — they fall through to EffectMutates and run automatically.

func IsReadOnlyCommand ¶ added in v0.6.3

func IsReadOnlyCommand(command string) bool

IsReadOnlyCommand reports whether every segment in the command is on the read-only allowlist and the command contains no mutation syntax (>, <, sed -i).

func Tools ¶

func Tools(workDir string, elicit *tool.Elicitation) []tool.Tool