What problem does it solve? Repeated permission re-approval prompts during agent sessions create toil and interrupt workflow, but blindly approving everything is unsafe. This Skill audits the re-approval log, grades each command pattern for safety, and recommends concrete configuration changes so recurring safe commands stop prompting and dangerous ones get denied. ## Core Features & Use Cases - Pattern Analysis: Reads the JSONL approval log, groups entries by tool and normalized command pattern, and counts occurrences per session. - A-F Safety Grading: Grades each pattern from read-only safe (A) to destructive (F) with justification covering blast radius, network access, and privilege level. - Actionable Recommendations: Suggests allow-list promotions, deny rules, narrower patterns, and alternative toil-reduction strategies, then applies approved changes via /update-config. - Use Case: After a long supervised session with dozens of pip install and docker build prompts, run the skill to get a stacked-card report grading each pattern, then promote safe ones to project settings and deny risky ones. ## Quick Start Ask the assistant to run an approval hygiene audit of this session's permission re-approvals and recommend settings changes.
