Skip to main content

Toolkit

Persistent Memory · Playbook

On this page7 sections

Playbook

1. Define Memory Classes

Separate durable preferences, repository discoveries, feedback, external references, and temporary project state. Give each class a retention expectation.

2. Require Provenance

Store origin, scope, observed or reviewed date, and confidence for facts that may drive action. Link to code, documentation, or an external record where possible.

3. Keep Team Rules Elsewhere

Put required behavior in version-controlled instructions or policy. Memory can route to that source; it should not become the only copy.

4. Review by Volatility

Expire active project state quickly. Review tooling and product behavior on a dated cadence. Retain stable preferences longer, with easy correction.

5. Test Retrieval

Verify what loads automatically, what is recalled on demand, how scope changes across repositories or worktrees, and what survives resume or compaction.

6. Provide User Control

Make memory inspectable, editable, exportable, and deletable. A correction should supersede or remove the old claim rather than append another contradiction.

7. Revalidate Before Consequential Action

Check current code, configuration, production state, or primary documentation. Memory accelerates discovery; it does not authorize release or external mutation.

Source register

Persistent Memory sources

Primary vendor documentation reviewed on the date shown. Links may change after publication.

  1. How Claude remembers your project (opens in a new tab)

    Anthropic · official docs · accessed 2026-08-01

  2. Custom instructions with AGENTS.md (opens in a new tab)

    OpenAI · official docs · accessed 2026-08-01

  3. Memories (opens in a new tab)

    OpenAI · official docs · accessed 2026-08-01

  4. GitHub Copilot CLI command reference (opens in a new tab)

    GitHub · official docs · accessed 2026-08-01