What problem does it solve? AI agents often over-ask for confirmation on obvious tasks, stall multi-step workflows waiting for approval, and let their memory stores bloat with stale or duplicated entries. This Skill acts as a parent router that directs incoming requests to two specialized sub-skills: one that decides when to act autonomously based on confidence thresholds, and one that keeps the agent's memory system clean and consolidated. ## Core Features & Use Cases - Autonomous Execution Threshold: Computes a dynamic confidence score from user input signals (sentence type, signal words, history) and applies a Predict-Judge-Act loop — execute silently at >=80% confidence, present options with predictions at 60-80%, and always confirm red-line actions like irreversible deletions. - Memory Optimization System: Offloads long tool outputs (>10KB or 50 lines) to context reference files with Mermaid summaries, runs FSRS-based memory health grading via a daily cron script, enforces tiered cleanup rules at 85%/90%/95% memory usage, and deduplicates memory against a fact store. - Use Case: A user says "continue" mid-pipeline. The router dispatches to the autonomous-execution-threshold sub-skill, which computes 0.95 confidence, executes the next pipeline step without asking, and appends a reasoning chain — while the memory sub-skill's cron job consolidates memory entries overnight. ## Quick Start Ask the agent to evaluate whether the current task meets the autonomous execution threshold, or request a memory consolidation and cleanup report for the current session.
