Skills Reference

Skills Reference

box-memory ships 12 skills. Each is invokable as /<skill-name> in Claude Code and Cowork; each also auto-fires from its description when the user asks something matching.

Quick reference table

Skill	Slash	Purpose	Tier
box-init	/box-init	Bootstrap a workspace (folder or Hub via --as-hub)	All; --as-hub Enterprise Plus
box-status	/box-status	Show workspace state, tier, counts, index health	All
box-tier-detect	/box-tier-detect	Probe Box capabilities, MCP variant, AI availability	All
box-mcp-check	/box-mcp-check	Verify official Box MCP at mcp.box.com	All
box-write	/box-write	Save a memory file with frontmatter + wikilinks	All
box-recall	/box-recall	Find memories via local _index.json	All
box-ai-recall	/box-ai-recall	Semantic Q&A via Box AI Ask (multi-doc or Hubs)	Business+ AI Units
box-companion	/box-companion	Describe a binary file (uses Box AI Extract on Business+)	All; AI path Business+
box-ai-extract	/box-ai-extract	Schema-driven metadata extraction via Box AI	Business+ AI Units
box-team	/box-team	Manage multi-team subtrees	All
box-ai-agent	/box-ai-agent	Persistent AI Studio "memory librarian"	Enterprise Advanced
box-index-rebuild	/box-index-rebuild	Refresh per-folder indexes from source	All

Skill categories

Setup & verification (use these first)

• box-mcp-check — confirm Box MCP is correct
• box-init — bootstrap your workspace
• box-tier-detect — see what your account can do
• box-status — sanity check the workspace

Memory CRUD (the everyday workflow)

• box-write — save
• box-recall — find (local)
• box-companion — describe a binary

Box AI features (opt-in, costs AI Units)

• box-ai-recall — semantic Q&A
• box-ai-extract — OCR + structured extraction
• box-ai-agent — persistent agent management (Enterprise Advanced)

Team / multi-tenant

• box-team — manage team subtrees

Maintenance

• box-index-rebuild — recover from drift

SKILL.md format

Every skill follows the same shape:

---
name: skill-name
description: <250-400 char third-person description with "Use when..." triggers>
argument-hint: "<expected argument shape>"
---

# /skill-name

> Box-MCP requirement / CONNECTORS reference

<1-paragraph what-this-does>

## Usage

<syntax + examples>

## What to do

<5-9 high-level steps>

## When to invoke / When NOT to invoke

## Errors to surface clearly

## Deep reference

<links back to repo docs>

This shape was tuned to pass Cowork's plugin validation. See Architecture for why each piece exists.

Opt-in AI flags

Box AI skills (box-ai-recall, box-ai-extract, box-ai-agent) and the AI path in box-companion are disabled by default. To opt in, set in _box-memory.json.settings:

settings:
  ai_recall_enabled: false       # set true for /box-ai-recall
  ai_extract_enabled: false      # set true for /box-companion AI path + /box-ai-extract
  ai_studio_agent_enabled: false # set true for /box-ai-agent
  ai_model: null                 # let Box pick (currently GPT-5 mini); override with specific model

See Box AI Units for the cost model.

Skill name format

All skill names: lowercase, kebab-case, prefix box-. The slash command IS the skill name with a / prepended — no separate "command" definition needed (modern Anthropic plugin format).

See also

• Quick Start — the minimum-useful sequence
• Schema — what the skills read and write
• Box AI Integration — depth on the AI-powered skills
• Operational Notes — Box quirks the skills work around