Knowledge capture. System context. Continuous improvement.

The Session Review.

A prompt by Anish, born out of the frustration of losing hard-earned context at the end of long, tiring AI pairing sessions.

The Vibe Coder

Anish was one of the first product owners to start working closely with Niraj and Sreenath, and he was the very first in the cohort to embrace "vibe coding."

What started as an experiment quickly evolved. Today, Anish is a full-blown AI Product Manager. He lives and breathes Claude, and he has committed himself to teaching his team—and his leaders—how to get the absolute best out of AI.

The Problem

When you spend a long Saturday pairing with an AI agent to build out a complex feature, you learn a lot. You discover non-obvious domain terminology, you establish conventions, and you figure out exactly how the system architecture should work.

But what happens when you close the laptop? All that context vanishes. The next time you open a new session, the AI has forgotten everything, and you have to start from scratch.

Anish realized that the end of a session is precisely when you have the most valuable context, but the least energy to manually document it.

The Solution

To solve this, Anish created his absolute favorite prompt: The Session Review.

You run this prompt at the very end of any meaningful AI brainstorming or working session. It asks the agent to look back over the entire chat history and extract the domain terms, non-obvious discoveries, and architectural decisions.

More importantly, it asks the agent to propose exactly where this knowledge should be saved (like CLAUDE.md or a .claude/skills/ file) so that future sessions can automatically inherit the context.

The Session Review Prompt
review this session and surface anything worth capturing for future sessions. look specifically for: - Domain Terminology - New or Non-Obvious knowledge discovered during this session you may include additional recommendations that you deem important For each item, propose: - A one-line summary of the learning - Where it belongs: CLAUDE.md (always-loaded project context), a SKILL.md under .claude/skills/ (procedure invoked on demand), or a slash command under .claude/commands/ (frequently-run action) - The exact content to write, in the format that destination expects Output the proposals as a numbered list first. Do not write or modify any files until I approve specific items by number.