add-memory

name	add-memory
description	Add a memory system to an agent — file awareness, knowledge graph, structured state, or multi-session tracking
argument-hint	[memory-type]
disable-model-invocation	false
user-invocable	true
allowed-tools	Read, Write, Bash, Glob, Grep, AskUserQuestion
metadata	{"version":"1.1","created":"2026-04-16T00:00:00.000Z","updated":"2026-05-16T00:00:00.000Z","author":"Ability.ai"}

Add Memory

Add a memory system to an existing agent. Memory skills are COPIED into the agent's .claude/skills/ directory — the agent becomes self-contained with no plugin dependency.

Memory Types

Type	Description	Skills Installed
file-index	File system awareness — creates and maintains a searchable file index	setup-index, refresh-index, search-files
brain	Zettelkasten-style knowledge graph — atomic notes with connections	setup-brain, create-note, search-brain, find-connections
json-state	Structured JSON state with jq-based updates and action logging	setup-memory, load-memory, update-memory, memory-jq
workspace	Multi-session project tracking — organized folders, sessions, archives	setup-projects, create-project, create-session, archive-project

Process

Step 1: Determine Agent Directory

Check if we're in an agent directory (has CLAUDE.md) or if user specified a path.

# Check current directory
if [ -f "CLAUDE.md" ]; then
  AGENT_DIR="$(pwd)"
else
  # Ask user for agent path
fi

If no CLAUDE.md found, ask:

• Question: "What agent should I add memory to?"
• Hint: "Enter the path to the agent directory (should contain CLAUDE.md)"

Step 2: Check Argument or Ask Memory Type

If user provided an argument (e.g., /agent-dev:add-memory brain), use that.

Otherwise, use AskUserQuestion:

• Question: "What kind of memory does this agent need?"

• Header: "Add Memory"

• Options:

  1. File awareness (file-index) — "I need the agent aware of workspace files"

     • Creates a searchable index of files in the workspace
     • Good for: agents that need to find and reference files

  2. Knowledge graph (brain) — "I need connected notes, evolving knowledge"

     • Zettelkasten-style atomic notes with connections
     • Good for: research agents, knowledge management, idea linking

  3. Structured state (json-state) — "I need counters, config, structured data"

     • JSON-based state with efficient jq updates
     • Good for: tracking metrics, configuration, action logs

  4. Multi-session workspace (workspace) — "I need organized projects and sessions"

     • Project folders with session tracking and archiving
     • Good for: long-running work, multiple parallel projects

Step 3: Copy Memory Skills

Memory templates are stored in this plugin at:

agent-dev/memory-templates/
  file-index/     # setup-index/, refresh-index/, search-files/
  brain/          # setup-brain/, create-note/, search-brain/, find-connections/
  json-state/     # setup-memory/, load-memory/, update-memory/, memory-jq/
  workspace/      # setup-projects/, create-project/, create-session/, archive-project/

Copy the selected memory type's skills into the agent:

# Get the plugin directory (where this skill lives)
PLUGIN_DIR="$(dirname "$(dirname "$(dirname "$0")")")"
TEMPLATES_DIR="$PLUGIN_DIR/memory-templates"

# Copy skills to agent
cp -r "$TEMPLATES_DIR/[selected-type]/"* "$AGENT_DIR/.claude/skills/"

Step 4: Update Agent CLAUDE.md

Add a section documenting the new memory capabilities:

## Memory: [Type Name]

This agent has [type] memory installed. Available skills:
- `/[skill-1]` — description
- `/[skill-2]` — description
...

Run `/[setup-skill]` to initialize.

Step 5: Confirm and Guide

Output:

## Memory Added

Installed **[type]** memory to `[agent-name]`.

### New Skills
- `/[skill-1]` — description
- `/[skill-2]` — description

### Next Step
Run `/[setup-skill]` to initialize the memory system.

Notes

• Skills are COPIED, not linked — agent becomes self-contained
• Multiple memory types can be installed in the same agent
• If skills already exist (duplicate names), warn and ask before overwriting
• The setup skill initializes the memory system (creates directories, indexes, etc.)
• Portability: all data directories are created inside the agent repo — never in ~/. If using the native Claude Code memory: frontmatter field on custom subagent definitions, use memory: project (stores in .claude/agent-memory/, stays in repo) — never memory: user (stores in ~/.claude/, breaks portability)