Claude Code Skills System

Give Your AI Agents Superpowers

Skills that make Claude Code agents proactively intelligent. Memory search, code graph queries, documentation lookup, and agent-to-agent messaging - all auto-triggered without you asking.

What Are Claude Code Skills?

Skills are SKILL.md files that teach Claude Code how to use specific tools and workflows. They're like plugins - but smarter.

📄

Simple Files

Just markdown files in ~/.claude/skills/. No compilation, no runtime.

🧠

Context-Aware

Claude reads skills and understands when and how to use them automatically.

Proactive

Skills auto-trigger based on context - no need to ask Claude to use them.

THE PROACTIVE PATTERN

Skills That Act Before You Ask

Three of our four skills follow a proactive pattern - they trigger automatically without you asking.

Example: You say "Fix the authentication bug in UserController"

1️⃣

memory-search

"What did we discuss about authentication?"

2️⃣

docs-search

"What's the documented behavior?"

3️⃣

Reads file

Opens UserController.ts

4️⃣

graph-query

"What depends on this?"

Result: Claude now has previous context, documentation, code structure, and dependencies - all before writing a single line of code.

🎯 Why This Matters

Without proactive skills, Claude might repeat explanations you've already heard, contradict previous decisions, miss dependencies that break, or start over instead of continuing where you left off.

Memory vs Planning: Two Different Problems

These skills complement each other but solve fundamentally different problems.

🧠

memory-search

Solves RETRIEVAL

"What did we discuss last week?"

  • 📅 Timescale: Days, weeks, months
  • 🔍 Problem: Can't recall past conversations
  • 💾 Data: Past session history (indexed JSONL)
  • Trigger: Proactive (auto-runs)
📋

planning

Solves EXECUTION

"What am I supposed to do next?"

  • ⏱️ Timescale: Minutes, hours (current session)
  • 🎯 Problem: Losing focus during complex tasks
  • 📝 Data: Current task files (task_plan.md)
  • 🔧 Trigger: On-demand (user invokes)

Use BOTH for complex work: Memory tells you what happened before. Planning keeps you focused now.

The Five Skills

Each skill gives Claude Code a specific superpower

🧠

memory-search

PROACTIVE

Solves RETRIEVAL: Search past conversation history from days/weeks/months ago. "What did we discuss last week?"

Auto-triggers when:

  • • User says "continue working on X"
  • • User mentions a previous decision
  • • User references "what we discussed"
  • • Starting any new task
# Example commands
memory-search.sh "authentication flow"
memory-search.sh "error handling" --mode semantic
📚

docs-search

PROACTIVE

Search auto-generated documentation for function signatures, class definitions, and API specs.

Auto-triggers when:

  • • User says "implement X function"
  • • User mentions a class/service
  • • Before calling unfamiliar functions
  • • Creating new components
# Example commands
docs-search.sh "UserController"
docs-find-by-type.sh function
🗺️

graph-query

PROACTIVE

Query the code graph database to understand relationships and impact of changes.

Auto-triggers when:

  • • After reading ANY code file
  • • Before modifying a function
  • • When changing a model
  • • Before refactoring
# Example commands
graph-describe.sh User
graph-find-callers.sh process_payment
graph-find-serializers.sh Order
💬

agent-messaging

ON-DEMAND

Send and receive messages between AI agents asynchronously. Multi-host support included.

Use when:

  • • Need work from another agent
  • • Completed a task for someone
  • • Something urgent happened
  • • Checking for requests
# Example commands
check-aimaestro-messages.sh
send-aimaestro-message.sh backend-api "Need API" "..."
📋

planning

ON-DEMAND

Solves EXECUTION: Stay focused during current complex tasks. Prevents goal drift, lost progress, and repeated errors within the current session.

Use when:

  • • Multi-step tasks (3+ steps)
  • • Research projects
  • • Building features
  • • Tasks requiring >5 tool calls

The 3-file pattern:

  • task_plan.md - Goals, phases, decisions
  • findings.md - Research, discoveries
  • progress.md - Session log, results
# Copy templates to project root
cat ~/.claude/skills/planning/templates/task_plan.md > task_plan.md
cat ~/.claude/skills/planning/templates/findings.md > findings.md

Installation

Get all five skills in under 2 minutes

1

Add the AI Maestro Marketplace

In Claude Code, run this command:

/plugin marketplace add 23blocks-OS/ai-maestro
2

Install the Plugin

/plugin install ai-maestro@ai-maestro-marketplace
3

Start Using Skills

Skills are immediately available! They auto-trigger when Claude Code agents read files or receive instructions.

That's it! No cloning, no manual installation. Skills work instantly.

⚠️

Service Dependency

4 of 5 skills require the AI Maestro service running on localhost:23000.

planning ✓ Works standalone
memory-search Needs AI Maestro running
docs-search Needs AI Maestro running
graph-query Needs AI Maestro running
agent-messaging Needs AI Maestro running

To enable all skills, install AI Maestro:

curl -fsSL https://raw.githubusercontent.com/23blocks-OS/ai-maestro/main/scripts/remote-install.sh | sh

Skills are available with the namespace:

ai-maestro:{memory-search, docs-search, graph-query, agent-messaging, planning}

Frequently Asked Questions

Can I use these skills with Aider, Cursor, or other AI agents?
The skills (SKILL.md files) are specific to Claude Code. However, the underlying command-line scripts (memory-search.sh, graph-describe.sh, etc.) work with any AI agent running in tmux. You can teach other agents to use these scripts directly.
Do I need AI Maestro running for skills to work?
4 of 5 skills require AI Maestro running on localhost:23000. These skills (memory-search, docs-search, graph-query, agent-messaging) communicate with AI Maestro's backend APIs.

The planning skill works standalone - it only creates local markdown files (task_plan.md, findings.md, progress.md) and doesn't need any backend service.
How do proactive skills know when to trigger?
The SKILL.md file contains trigger conditions that Claude Code understands. For example, memory-search triggers when you mention "continue" or reference previous work. Claude reads these conditions and automatically invokes the skill before starting work.
Can I create my own skills?
Absolutely! Skills are just markdown files. Create a new folder in ~/.claude/skills/your-skill/ with a SKILL.md file. Include frontmatter with name, description, and allowed-tools. Then document when the skill should trigger and what commands to run.

Ready to Give Your Agents Superpowers?

Install AI Maestro Skills and watch Claude Code become proactively intelligent.