memory

name	memory
description	Persist and recall facts, config, and project state. Use when the user says "remember this", "save this", "where does X go", "archive this project", "add this to daily notes", or any operation involving storing or retrieving persistent state. For credentials (API keys, tokens, passwords), route to the `secrets` tool in the parent `core` skill instead.
allowed-tools	["Read","Write","Edit","Bash","Glob","Grep","AskUserQuestion"]

Memory

Get, set, and save operations on persistent state. All operations happen within a scope.

Secrets/credentials are not handled here — see the secrets section of the parent core skill, which routes to the Karelin Keys MCP (Azure Key Vault).

Scopes

Scope	Storage backends	Examples
my	Obsidian vault	daily notes, personal config
team	shared repos, shared drives	team config
project	project repo, project docs	project state, project config
company	company systems	company-wide config

Scope is detected from context or elicited from the user.

Operations

get

Recall a fact or piece of state.

• Facts: Obsidian vault, AGENTS.md, auto-memory
• Config: project files, AGENTS.md locations

set

Persist a fact or piece of state to the correct location.

1. Identify the entity type (location, config, fact, mistake, etc.)
2. Determine scope (my/team/project/company)
3. Route to the correct backend based on entity type + scope
4. Persist

Known entity → backend mappings:

• Locations → AGENTS.md Locations section
• Daily notes / mistakes → Obsidian vault daily notes
• Project facts → project docs or AGENTS.md
• Skill metadata → SKILL.md frontmatter

save (export/archive)

Bundle a project folder's complete state for handover or archival:

1. Extract git diff history for the folder
2. Find related .claude conversation logs
3. Copy current source snapshot
4. Generate README summary
5. Create .tar.gz archive
6. Optionally push to a target repo

Uses: scripts/extract-claude-history.py

Entity Recognition

Before executing any operation, identify all referenced entities. If any are ambiguous or unrecognized, collect all uncertainties and ask in one batch before proceeding.