learn-from-chat

What problem does it solve?

This Skill ensures insights that emerge during agent-user conversations are captured, validated, and turned into safe, targeted updates to skills or processes so documentation and behaviors stay accurate and current.

Core Features & Use Cases

• Capture & Evidence: Extract a one-line insight and the concrete chat evidence that proves it (bug, workaround, missing guardrail).
• Classification & Matching: Classify the learning (GOTCHA, TECHNIQUE, FAILURE_MODE, METRIC, CONTRADICTION) and locate affected SKILL.md files.
• Propose, Validate & Log: Present exact diff-style change proposals for user approval, run validation checks, bump versions, and append a dated entry to a shared research log.
• Use Case: During a debugging session, automatically surface a repeated failure mode, propose a guardrail addition to the affected skill, and log the change after explicit user approval.

Quick Start

Capture a discovered issue by asking the agent to summarize the insight, provide evidence, classify it, and propose exact SKILL.md changes for approval.