📡 Spellbook

A library of reusable agent “spells.” Supercharge your local AI with structured prompts, advanced automation routines, and repeatable workflow templates.

Overview

Spellbook is a prompt and automation library for the Companion Intelligence ecosystem. Think of it as a collection of reusable “spells” — structured prompt templates, multi-step agent workflows, and automation routines — that you can invoke to get consistent, high-quality outputs from your local LLMs.

Instead of writing the same prompts repeatedly or keeping them scattered in notes, Spellbook provides a centralized, versioned, shareable library that integrates directly with your Hub’s agent runtime.

Key Features

• Prompt library — organized collection of role-based and task-specific prompts
• Multi-step workflows — chain agent calls, tool invocations, and memory queries into repeatable sequences
• Variables & parameters — define inputs for each spell so they are reusable with different data
• Triggers — run spells on schedule, on Hub events, or via voice command
• Community spells — import spells from the CI Marketplace
• Version control — full history of every spell edit with rollback support
• MCP integration — expose spells as MCP tools callable by any connected agent

Use Cases

• Daily briefing spell: “Summarize my Digital Memory from yesterday, check the weather, list today’s tasks”
• Research spell: “Given this article [clip], write a structured summary with key points and counter-arguments”
• Meeting prep spell: “Review my memory for everything related to [project] in the past week”
• Code review spell: “Review this diff against our style guide and list issues by severity”
• Survival drill: scheduled weekly preparedness checklist (see Just In Case)

Spell Format

Spells are defined in YAML:

name: daily-briefing
description: Morning summary of memory, tasks, and schedule
version: 1.0.0
tags: [productivity, daily]
 
inputs:
  - name: days_back
    type: integer
    default: 1
    description: How many days of memory to summarize
 
steps:
  - id: memory_summary
    tool: digital_memory.search
    params:
      query: "activities and notes"
      days: "{{inputs.days_back}}"
 
  - id: briefing
    tool: llm.complete
    params:
      model: default
      prompt: |
        Given the following memories from the past {{inputs.days_back}} day(s):
        {{steps.memory_summary.result}}
        
        Write a concise briefing covering:
        1. Key activities and conversations
        2. Open tasks or follow-ups
        3. Anything that needs attention today
 
output: "{{steps.briefing.result}}"

Setup

Usage

Running a Spell

In the Spellbook UI, open any spell, fill in its inputs, and click Run. Output appears in the results panel.

Scheduling a Spell

Open a spell → Schedule → set a cron expression or natural language schedule (“every weekday at 8am”).

Exposing as MCP Tool

Open a spell → Settings → Expose as MCP Tool. The spell is now callable by any Companion Agent that has Spellbook’s MCP endpoint in its server list.

CLI Usage

ci-spellbook run daily-briefing --days-back 2
 
# List all spells
ci-spellbook list
 
# Import from Marketplace
ci-spellbook import marketplace://spellbook/research-assistant

Troubleshooting

Spell produces inconsistent results Add explicit formatting instructions and few-shot examples to the prompt step. Use temperature: 0.2 for deterministic tasks.

MCP tool not visible to agents Check that Spellbook is running and the MCP endpoint (http://spellbook.ci.localhost/mcp) is configured in your agent’s MCP server list.

Scheduled spell not running Check the scheduler logs in Spellbook → Logs → Scheduler. Ensure the Hub’s timezone is configured correctly under Hub → Settings → System.

Links

• GitHub Repository 
• Digital Memory Server
• Just In Case