memory packs

Pre-built knowledge and skills for memory engine — import a YAML file, get instant value.

What are memory packs?

A memory pack is a YAML file containing memories that you import with me pack import. No code, no plugins, no running processes.

Packs may contain a mix of:

• Reference knowledge — static content that's useful immediately after import
• Skill instructions — procedural content that teaches AI agents how to perform tasks

Memory Packs are Neo's "I know kung fu" for your agents.

Quick start

# Validate a pack (no server needed)
me pack validate packs/git-history.yaml

# Import a pack
me pack import packs/git-history.yaml

# Preview what would happen
me pack import --dry-run packs/git-history.yaml

Available packs

Pack	Description
pack-authoring	How to write effective memory packs — skill memory structure, reference memory craft, organization, ID/versioning, and quality checklist
skill-to-pack	Convert Agent Skills (agentskills.io) into memory engine packs — format reference, conversion procedure, field mapping, and edge cases
codebase-index	Teaches agents to build a structural code index using Tree-sitter and filesystem watching — zero LLM cost, always current
docs-to-pack	Convert project documentation (Markdown, MDX, RST, OpenAPI) into memory packs
git-history	Teaches agents to backfill and incrementally sync git commit history into searchable memories
github-content	Extract GitHub issues, PRs, and releases into memory engine using the gh CLI
build-postgres-queue	Comprehensive guide for designing and implementing task queues using PostgreSQL

Pack format

Each pack is a YAML file with a top-level envelope declaring pack identity, followed by a memories array:

name: example_pack
version: "0.1.0"
description: "One-line summary of what this pack provides"
id-prefix: "019b0000"
format: 1
memories:
  - id: "019b0000-0001-7000-8000-000000000001"
    tree: "subtopic"
    meta:
      type: "skill"
    content: |
      # Example Skill — What This Does

      Instructions for an agent...

Envelope fields

Field	Required	Purpose
name	Yes	ltree-safe identifier ([a-z0-9_]+) — used to construct tree paths at import
version	Yes	Semantic version string (e.g., "0.1.0")
description	No	Human-readable summary
id-prefix	Yes	8 lowercase hex characters — all memory IDs must start with this
format	Yes	Must be 1
memories	Yes	Array of memory objects

Conventions

• Envelope: declares pack identity (name, version) — individual memories do not include meta.pack
• IDs: deterministic UUIDv7 with a fixed prefix per pack (makes re-imports idempotent)
• Tree: relative paths — auto-prefixed with pack.<name>. at import time
• Content: non-empty, self-contained, follows memory best practices

Additional meta keys (like type, topic, etc.) are optional — add whatever makes your content discoverable and filterable.

ID allocation

Each pack claims a unique UUIDv7 prefix. Declare it in the envelope's id-prefix field.

Prefix	Pack
019b0001	pack-authoring
019b0002	skill-to-pack
019b0003	codebase-index
019b0004	docs-to-pack
019b0005	git-history
019b0006	github-content
019b0007	build-postgres-queue

When adding a new pack, choose the next available prefix.

To generate IDs for a new pack:

bun scripts/pack-uuids.ts 019b0007        # 10 IDs (default)
bun scripts/pack-uuids.ts 019b0007 25     # 25 IDs
bun scripts/pack-uuids.ts 019b0007 5 3    # 5 IDs starting at sequence 3

Upgrading packs

me pack import handles version-aware upgrades automatically:

# First import: installs v0.1.0
me pack import packs/git-history.yaml
# Imported 1 memory, deleted 0 (pack: git_history@0.1.0)

# After updating the file to v0.2.0:
me pack import packs/git-history.yaml
# Imported 2 memories, deleted 1 (pack: git_history@0.2.0)

Old-version memories are automatically cleaned up. Memories that exist in both versions are updated in place (deterministic IDs).

Contributing

1. Fork this repository
2. Create your pack as a YAML file in packs/
3. Validate with me pack validate packs/your-pack.yaml
4. Submit a pull request

All packs must:

• Use the v2 envelope format (name, version, id-prefix, format: 1, memories)
• Use an ltree-safe pack name ([a-z0-9_]+ — no hyphens)
• Use a unique ID prefix (check the table above)
• Have non-empty, well-written content

Validation & CI

Every pack is validated on push and pull request via GitHub Actions. The workflow checks:

1. Per-pack validation — envelope fields, schema, IDs match id-prefix
2. Cross-pack duplicate names — no two packs may share a name
3. Cross-pack ID prefix collisions — no two packs may share an ID prefix

Local validation

# Single pack
me pack validate packs/git-history.yaml

# All packs (same checks as CI)
./scripts/validate.sh

The script requires me and yq on PATH.

License

MIT