Agents

This Skill addresses common issues in AI Agent runtime, such as infinite loops, repeated tool calls, deviation from original instructions after multi-turn dialogues, hallucinated tool parameters, premature task termination, and silent suppression of tool errors. It covers core dimensions of Agents including loop control, tools, memory, context budget, cost, and upgrades. Framework-agnostic, it allows you to focus on diagnosing and optimizing Agent behavior.

Usage is straightforward: simply describe the problem in natural language, e.g., "My Agent is stuck in a loop" or "Help me analyze token consumption." The Skill automatically reads local configuration and memory files, analyzes the Agent's current state, and provides specific diagnostic results and remediation suggestions. You can also use it to design tool schemas, retry strategies, timeout settings, checkpoints, human approval rules, and to write evaluation test suites and regression suites.

It is ideal for Agent developers, operations engineers, and project managers who frequently debug and optimize AI Agents—especially teams or individuals who have deployed multiple Agents and need unified management of loop control, memory design, multi-Agent switching, tracking, canary releases, as well as control over token costs and latency.

It is recommended to prioritize using this Skill for systematic diagnosis whenever Agent behavior is abnormal, rather than blindly modifying prompts. Note that it only processes local files (notes in the ~/Clawic/data/ directory), does not send any data externally, and does not write credentials. Ensure you have created basic files such as config.yaml and memory.md as per the documentation; otherwise, it will start with default values.