setup-memory

name

setup-memory

description

Set up version-controlled, per-project Claude memory that lives IN the repo (memories/repo/) and is auto-loaded via a filesystem junction at ~/.claude/projects/<slug>/memory. Use when starting Claude on a new project, when the user asks to "set up memory", "memory in repo", "share memory with the team", "make memory reusable across projects", or to migrate an existing ~/.claude project-memory folder into the repo. Generic + publishable — no secrets, no per-user content.

Setup Memory — memory-in-repo

Claude's per-project memory normally lives outside the repo at ~/.claude/projects/<slug>/memory/, so it is never versioned or shared. This skill moves it into the repo at memories/repo/ and links it back with a directory junction, so the curated project knowledge is:

• Committed + team-shared — clone the repo, run the setup, and Claude instantly has the project's context.
• Token-efficient — Claude reads a tight, hand-maintained MEMORY.md index instead of re-discovering the codebase every session. That is the whole point: pay once to write a memory, save tokens on every future session.
• Just files — adding/updating memory is a normal file edit in the repo (the junction makes it live immediately).

Each project keeps its own memory (per-project; nothing is shared between projects except this generic skill).

Run

& "$HOME\.claude\skills\setup-memory\scripts\Setup-Memory.ps1" -ProjectPath <repo-root>
# -ProjectPath defaults to the current directory. Add -WhatIf to preview, -Force to re-seed templates.

Idempotent and no admin required (a directory junction, not a symlink). The script:

1. Computes the project slug from the path the same way the harness does: lowercase the drive letter, then turn : \ / . _ and spaces into - (e.g. C:\Projects\Foo.Bar_v2 → c--Projects-Foo-Bar-v2).
2. Creates <repo>/memories/repo/ and a MEMORY.md index skeleton (Feedback / Reference / Project / User) if missing.
3. If ~/.claude/projects/<slug>/memory already exists as a real folder, migrates its files into memories/repo/ first (never overwrites a newer file), then removes the empty folder.
4. Replaces it with a junction: ~/.claude/projects/<slug>/memory → <repo>/memories/repo.
5. Seeds _TEMPLATE.memory.md (frontmatter template) + a README.md if missing. Prints slug, paths, and migrated-file count.

Memory file convention (what goes in memories/repo/)

• One fact per file, named <type>_<kebab-slug>.md where type ∈ feedback | project | reference | user.
• Frontmatter:

  ---
  name: <type>_<kebab-slug>
  description: <one line — used for recall relevance>
  metadata:
    node_type: memory
    type: feedback | project | reference | user
  ---
  <the fact. For feedback/project, add **Why:** and **How to apply:** lines. Cross-link with [[other_name]].>
  

• MEMORY.md is the index loaded every session: one bullet per file under its section, - [file](file.md) — hook. Never put memory content in MEMORY.md.
• type meanings: user = who the user is / preferences · feedback = how Claude should work here (with the why) · project = ongoing work/goals/constraints not derivable from code or git · reference = pointers to external resources (URLs, dashboards, tickets).

Notes

• Commit memories/ to share with the team. Keep it public-safe — no secrets, tokens, or credentials.
• Re-run any time; safe and idempotent. Use -WhatIf to see what it would change.
• Generic by design (no hardcoded user, org, or project names) so it can be published and reused anywhere.