Claude Code project memory

Persistent project memory for Claude Code.

Give Claude Code durable product truth, task state, branch lanes, claims, and audit trail across sessions with RAAV.

What RAAV gives agents

Claude Code reads the same product handbook on every run
Onboarding prompts label facts as confirmed, inferred, or needing founder review
Task claims prevent parallel agents from editing the same files blindly
Submits record files, summary, verification, risks, and memory candidates

The problem

Claude Code can ship useful work in one session, but product context, founder decisions, and verification status often disappear into chat scrollback.

The RAAV layer

RAAV gives Claude Code a persistent project memory it can read before planning and update after work, while keeping the founder in control of durable product truth.

Claude Code is stronger when product context survives the session

Claude Code can reason deeply about a repository, but a founder still needs continuity. The important product facts, open decisions, branch ownership, and verification expectations should not depend on a single chat transcript.

RAAV gives Claude Code a product briefing before it plans

A good Claude Code session starts with the right context. RAAV gives the agent a Product Handbook, current task plan, active lanes, accepted requirements, risks, and unresolved founder questions before it decides what to edit.

That context is repo-local and repeatable. A new Claude Code session can read the same operating memory instead of asking the founder to explain the product again.

Confirmed facts are separated from inferred assumptions
Open founder questions are visible before implementation
The next task can be selected from current project state

Claude Code can propose product changes without rewriting truth

Coding agents often discover product gaps while implementing. Maybe an acceptance criterion is missing, a task is too broad, or a requirement conflicts with the current UI. Those discoveries are valuable, but they should not silently become permanent memory.

RAAV routes those findings through proposals and memory candidates. Claude Code can suggest a requirement, task, decision, or durable memory update. The founder confirms it before it becomes shared truth for every future agent.

Propose new requirements or task splits
Attach memory candidates during submit
Confirm, defer, reject, or reopen proposals later

Handoffs become useful after the agent stops

The end of an agent session is where many projects lose clarity. A chat summary can be helpful, but it rarely ties files, verification, risks, and next actions into a durable project record.

With RAAV, Claude Code submits a structured result: what changed, which files were touched, what verification passed, what risks remain, and which facts should be reviewed for memory. The next session starts from that trail.

Files and task state stay attached
Verification is recorded explicitly
The founder can review work without reading the entire chat

How the agent loop works

RAAV does not replace your coding agent. It gives the agent a durable operating system for product truth, task ownership, and proof.

Install

Paste the Claude Code prompt from RAAV install and run `raav init` if needed.

Onboard

Run the onboard-and-plan workflow so Claude Code drafts product memory and backlog proposals.

Work

Use `raav pack work`, `raav next`, `raav claim`, and conflict checks before editing.

Handoff

Submit results and promote only founder-approved memory.

Claude Code install prompt

The founder does not need to become a CLI expert first. Paste a direct instruction into Claude Code and let the agent install RAAV, inspect the repo, and prepare the first product memory pass.

Paste this into your coding agent

Install RAAV for this repository. Use it as persistent product memory for Claude Code. Create the Product Handbook, list inferred assumptions separately from confirmed facts, propose missing requirements, and show me the first task before editing.

After setup, Claude Code should use RAAV commands before planning, claiming files, and submitting results.

Why this is different

Most tools either write code or track human tickets. RAAV sits between the founder and the coding agents as shared product memory, coordination, and audit.

Works through CLI and MCP-friendly workflows

Keeps reasoning inside Claude Code rather than charging for another model layer

Founder review queue separates assumptions from confirmed product truth

Claude Code with chat memory vs Claude Code with RAAV

Session start

The founder repeats goals, constraints, and recent decisions.

Claude Code reads the Product Handbook and current context pack.

Parallel work

Another agent may be editing the same files without visibility.

Claims and lanes show active work before Claude Code edits.

Assumptions

A guessed requirement can slip into the implementation.

Assumptions become proposals or memory candidates for founder review.

After the run

The summary lives in chat and may be hard to reconstruct later.

RAAV records files, summary, verification, risks, and next actions.

FAQ

Claude Code users usually care about context continuity, MCP workflows, and avoiding repeated product explanations.

Does RAAV replace Claude Code memory?

No. RAAV gives Claude Code a repo-local product memory and workflow protocol. Claude Code still reasons over the codebase. RAAV makes the product state durable and reviewable across sessions.

Can Claude Code use RAAV without hosted sync?

Yes. The core RAAV flow is local-first. Claude Code can use CLI and MCP-style workflows to read context, claim tasks, submit results, and queue memory candidates without requiring a hosted service.

What should Claude Code write into memory?

Confirmed product facts can become durable memory. Inferred facts should be submitted as candidates or proposals so the founder can approve, defer, or reject them.

Is this mainly for solo founders?

Solo founders are the sharp first use case because they run many agent sessions themselves. The same memory layer also helps small teams using Claude Code alongside Codex, Cursor, or other agents.

Related RAAV workflows

Claude Code install prompt Product Handbook Claude Code guide