Agent Metacognition & Planning

•
54
•

Enhances AI agent autonomy through self-correction, strategic planning, and adaptive context management.

The Meta skill implements a robust self-improvement and metacognition layer for Claude Code, enabling agents to 'think about thinking' before executing tasks. It provides a structured framework for session orientation, pace control, and error prevention by leveraging persistent memory systems like lesson logs and context maps. By enforcing deliberate planning phases and systematic self-checks, this skill prevents agent spiraling and ensures high-quality, non-generic output in complex development environments.

Key Features

01Intelligent Pace Control: Dynamically adjusts execution speed based on task complexity and uncertainty.
02Persistent Learning Loop: Logs and promotes recurring mistakes to a permanent LESSONS database for future sessions.
03Context-Aware Navigation: Prioritizes MAP and Journal lookups to ensure the agent understands the project structure.
04Self-Correction Triggers: Automatically pauses execution when detecting generic patterns, scope creep, or repetitive errors.
05Skill Composition: Stacks seamlessly with domain-specific skills like backend or design for more deliberate execution.
0654 GitHub stars

Use Cases

01Onboarding to large, complex codebases where initial orientation and context mapping are critical.
02Executing high-stakes refactoring that requires a strict 'plan-first, code-later' methodology.
03Recovering from repetitive logic loops or errors during long-running autonomous tasks.