safe-autonomous-mode

What problem does it solve?

Autonomous agents and unattended workflows can take unsafe or irreversible actions without clear, project-aware guardrails; this Skill provides structured risk classification and allow/deny policy composition to prevent dangerous or unexpected changes during hands-off operation.

Core Features & Use Cases

• Per-action risk classification: Evaluate every proposed action with a structured verdict including reasoning and a binary allow/block decision.
• Allow/Deny rule composition: Maintain conservative defaults for safe behavior while permitting project-specific overrides for read-only tasks, builds, and non-destructive edits.
• Pacing and focus-aware autonomy: Tick-based execution, adaptive sleep durations, and different behavior when the user is focused versus away to avoid token waste and unnecessary prompts.
• Use Case: Run an autonomous code maintenance agent that reads tests, runs builds, commits non-destructive changes, and requests confirmation for pushes, destructive shell commands, or database mutations.

Quick Start

Ask the agent to run in safe-autonomous-mode and classify each proposed shell, git, and file-modifying action before execution so it blocks ambiguous or high-risk operations.