Agent Metacognition

Enables agents to self-monitor execution, reflect on tasks, analyze errors, and adjust learning strategies through metacognitive processes.

This package appears to be a self-contained, in-process metacognition library (monitoring, reflection, experience store) and does not request secrets or perform network calls. Before installing: (1) run it in a sandboxed/test agent to confirm behavior; (2) verify the example usage is adapted to your agent (the SKILL.md references an 'agent' object that must be provided by your environment); (3) note that experiences and reflections are stored in memory (no persistence) so data is ephemeral unless you modify the code; (4) review the full source if you plan to use it in production—look for any future changes that might add I/O, network, or credential access. The package author/source is listed as unknown, so use usual caution when running third-party code.

Type: OpenClaw Skill Name: yuyonghao-agent-metacognition Version: 0.1.0 The skill bundle implements a metacognition system for tracking agent performance, including self-monitoring and reflection logic. The code (src/reflection.js, src/self-monitor.js) operates entirely in-memory, using standard Node.js patterns to record task outcomes, calculate confidence scores, and identify performance patterns without any external network calls, file system access, or sensitive data exposure.