What problem does it solve? Claude Code normally prompts for approval on every tool call, which blocks unattended or long-running work sessions. This Skill lets you enable a bounded autonomic work mode that auto-approves tool calls for a specific project and time window, while keeping destructive commands, network egress, and out-of-project writes gated behind normal permission prompts. ## Core Features & Use Cases - Two auto-approval modes: partner mode (default 8h) auto-approves tool calls while leaving questions to the user; away mode (default 4h) additionally denies AskUserQuestion since no one is available to answer. - Three enforced guards: wall-clock expiry capped at 12 hours, per-project scoping so one repo's window never arms another, and a denylist blocking destructive shell commands, WebFetch/WebSearch, and writes outside the project. - Audit trail: every auto-approval, denial, mode change, and registered judgment call is logged to an SQLite-backed decision log with 60-day retention. - Use Case: Before leaving for the evening, run /auto-wm away 4h in your repo so Claude continues a refactoring task unattended, auto-approving safe edits, skipping questions, and logging every decision for morning review. ## Quick Start Ask Claude to enable autonomic work mode by saying "enable auto-wm away mode for 4 hours" and it will run the awm.py script, verify the hook fires, and confirm the window and scope.