@agentic.james: My Claude Code actually improves its own skills overnight, here's how. Every conversation you have with Claude Code, including every tool call and its thinking, gets stored in a JSONL transcript in your Claude folder. That's a complete paper trail of how it behaves, and almost nobody uses it. Since skills get invoked by tool calls, you can look in that transcript to see exactly how a skill changed the agent's behavior and whether it followed the instructions. Once you can see that, you can run an auto-research loop on it: one agent grades another agent's transcript after it used a skill, figures out what went wrong, and rewrites the skill so the next run is better. That's how your skills get sharper while you sleep. Link in bio.

agentic.james

Open In TikTok:

Region: US

Thursday 11 June 2026 15:15:34 GMT