The ask command is an intended AI assistant feature, but its one-shot Claude path silently enables automatic shell-tool execution despite the interactive mode offering permission prompts. This creates a meaningful risk of unintended command execution, so the package should receive a warning rather than a malicious verdict.
Static reason
One or more suspicious static signals were detected.
Trigger
The user invokes `smbls ask` with a question while the Claude provider is configured.
Impact
Commands returned by the AI can act with the CLI process's privileges and may read or modify project files or perform other shell actions without a per-command prompt.
Mechanism
The one-shot Claude path sets permission mode to automatic. The assistant can invoke `run_command`, which executes its command in the current working directory.
Evidence
dist/esm/bin/ask.jscurrent working directory (command-dependent)