Recall

Overview

Recall is a local-first CLI that indexes AI coding sessions from multiple tools. Treat it as a project memory layer: use it to recover history, reduce repeated mistakes, and turn prior sessions into current review evidence.

Use the installed recall command when inspecting the user's session history. If you are developing Recall itself, use the project build only for testing Recall behavior, not as a substitute for the user's installed history index.

If recall is unavailable, stop the Recall workflow, tell the user it is not installed, and offer brew install samzong/tap/recall instead of pretending history was inspected.

Recall history is evidence, not truth. Verify technical claims, code behavior, commands, paths, and invariants against the current repository before acting.

Scoping Protocol

When the request is broad, such as "use Recall to review this project", do not jump straight to full export. Ask one concise scoping question that makes Recall's value clear:

Default Recall review: historical risk audit, decision archaeology, failed approaches, unfinished work, and current-code assumptions to verify. Should I start with that quick scan, go deep across all project history, or focus on a topic like architecture, tests, performance, a feature, or PR readiness?