crew-memory

Organizes durable project memory in an Obsidian vault with anchored, index-first markdown notes.

2|Updated Jul 28, 2026
One-click install
npx skills add https://github.com/mbadali25/useful-claude-add-ons --skill crew-memory-mbadali25
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: crew-memory
Source: https://github.com/mbadali25/useful-claude-add-ons/tree/main/plugin/crew/skills/crew-memory
Command: npx skills add https://github.com/mbadali25/useful-claude-add-ons --skill crew-memory-mbadali25

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Long-running projects across multiple repositories lose context: subsystem maps, architectural decisions, and cross-repo API contracts live nowhere durable, and agents re-derive them expensively or trust stale notes. This Skill defines conventions for storing that knowledge in an Obsidian vault without letting unbounded notes blow up the context window. ## Core Features & Use Cases - Vault layout conventions: Standardizes paths for repo codemaps, decision records, and cross-repo contracts, plus a single INDEX.md entry point. - Index-first reading rule: Only INDEX.md is read by default; all other notes are fetched by path on demand to keep token usage bounded. - Anchor-based staleness detection: Every note carries an anchor: <repo>@<sha> so git diff can reveal when referenced files moved and the note needs re-verification. - Use Case: A team with 5+ repos records that service A's endpoint is consumed by service B in a way neither repo's code shows; the contract note in contracts/ captures it, and anchors catch when it drifts. ## Quick Start Set up an Obsidian vault for project memory with an index, codemap mirrors, and anchored decision notes following the crew-memory conventions.

Frequently Asked Questions about crew-memory

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I set up an Obsidian vault for project documentation?

Create folders for repos/<repo>/codemap, repos/<repo>/decisions, and contracts, plus a single INDEX.md with one line per note. Obsidian needs no integration layer since it is plain markdown that Claude Code reads and writes directly.

How do I keep AI agents from reading too many notes?

Use index-first reading: only INDEX.md is loaded by default, and every other note is read by path on demand, one at a time. If a task needs more than three notes, the notes are badly organized and should be restructured.

How do I detect stale documentation notes?

Give each note an anchor field with the repo and commit SHA it was written against, then run git diff --name-only <anchor-sha>..HEAD on the referenced paths. If those files changed, re-verify that section before trusting it.

What should I do when a note contradicts the code?

Code always wins over notes. When they disagree, the note is wrong; fix the note rather than reasoning from it, because stale notes are confidently wrong in ways fresh code searches are not.

What content belongs in a project memory vault?

Store subsystem maps, cross-repo contracts, decisions with rejected alternatives, and failed experiments. Exclude anything derivable from code in under a minute, API docs, or ticket copies.

How do I sync repo codemaps with an Obsidian vault?

Treat the repo-local .crew/codemap directory as the source of truth and symlink it into the vault when both are on the same machine. Symlinking avoids copies diverging over time.