introspect
Introspect
Turn what happened in your AI coding threads into changes to how your agents work for you — so your setup compounds instead of resetting every session.
It's agent-agnostic. It reads your threads from Claude Code, OpenAI Codex, and Cursor,
finds the handful of things worth changing, and — with your approval — writes them back into
whichever agent's config actually steers future sessions. Claude Code already ships /insights,
which reads your last 30 days and shows you a report. That's the read half, Claude-only.
Introspect is the write half, across every agent you use. Youssef put it well in the thread
this came from: fix the system once so you don't have to fix the output a thousand times.
First progress marker
Start your first progress update with the words Using introspect. Keep the phrase stable — since
this skill reads transcripts to find recurring patterns, a stable marker lets a later run (or the
skill's own reliability review) see when it was invoked and whether it helped.