Learn From Your Own Record
Point the agent at the record of past sessions and ask what it shows. The evidence is already on the disk.
The prompt
"Research your own experience. Absorb what is useful, reject what is useless" (@ethanolivertroy, after Bruce Lee)
The longer version treats session archives as an honest record and walks the agent through six gated phases, from finding the archives to rebuilding the user's tools with consent. Source: X/@EXM7777, 2026-08-09
Lauren Tan's version is narrower: have the agent read past review comments and transcripts to learn how the user steers agents, then find the pitfalls agents hit most. Source: X/@poteto, 2026-10-16
Why it works
Transcripts replace self-report with counts: a correction that recurs across many sessions is a fact about how you work, not an impression. Transcripts record what was actually asked and done, and an agent can read thousands of them. The gates in the long version (evidence before verdict, interview before conclusions, consent before changes) stop it from turning a few sessions into a confident story.
When to use it
- When the same correction keeps coming up in different sessions.
- As a periodic review of where time goes.
Session archives hold private material. Run this locally, keep the evidence file out of any repo, and read what the agent proposes to change before it changes it.
Which skill runs it
learn-from-projects mines recent transcripts across Kevin's projects into candidate wiki learnings behind a codify gate. The wiki's Wiki State Machines page notes that this harvest has no scheduled executor yet, so it runs only when someone starts it.
Let agents report friction
Steve Ruiz gave agents a tiny "papercuts" CLI to complain about dead-end tool calls, broken links and other friction they hit during work. Without it, models push through and never mention the problems. The agent's own record of friction is a list of fixes for the environment. Source: X/@steveruizok, 2026-07-09
A one-line version for coding: ask the agent to analyze your own usage habits and coach you to work with it better. Source: X/@hzlzh, 2026-07-20
Timeline
- 2026-11-03 | Page created from three prompts that point agents at their own session history. Source: X bookmarks, 2026-07-20 to 2026-10-16