Skills are created with YAML frontmatter by default:
---name: skill-name
description: Provides [what this skill does]. Use when [trigger context].---
The description field should be written in third person — describe what knowledge the skill provides and when an AI agent should load it. Example: "Provides the company refund and returns policy. Use when handling customer queries about refunds or exchanges."
Tag commands
Tags are flat and untyped. A tag exists only as a #value token inside a file's content, and Contextium reconciles a file's tags from its body when the file is saved.
Command
Description
contextium tags list -w <workspace>
List all tags in a workspace
contextium tags search "<query>" -w <workspace>
Search files by tags
contextium tags file-tags <fileId> -w <workspace>
List tags applied to a specific file
Tags are entirely content-driven — these are read-only commands. There is no tags create, tags update, or tags delete: a tag exists only while some file's content contains its #value, and it appears/disappears as files are edited. Tag a file by adding a #value token to its content (e.g. contextium edit-file <id> --content "… #urgent"); Contextium auto-creates the tag on save and reconciles a file's tags from its body. To untag, remove that token. contextium tags apply, tags apply-bulk, and tags remove are deprecated — they print guidance to author tags inline rather than applying or removing them.
Comma-separated member names (default: everyone in the workspace)
--email <emails>
Comma-separated email addresses
--member-id <ids>
Comma-separated member IDs
-m, --message <message>
Custom message to include in the notification
If a name matches more than one member, the command fails and lists all matches — re-run with --email to be precise. Notifications respect each member's preferences and Do Not Disturb settings.
Marketplace commands
Command
Description
contextium marketplace list --query "<description>"
When the Contextium CLI or MCP server is connected inside Claude Code, a set of /ium: slash commands become available. These commands orchestrate multi-step Contextium workflows directly from the Claude Code chat window — no terminal required.
Getting started
Command
Description
/ium:code-audit
Audit an existing codebase for quality and risk — security observations, tech debt, fragile/high-risk areas, test-coverage gaps, and dependency risks — and write a severity-rated audit report to a Contextium library. Use /ium:map-existing instead when you want to document structure and plan a Contextium migration.
/ium:new-project
Set up a new Contextium project from scratch — walks you through workspace selection, then intelligently suggests and creates context libraries, agents, skills, tags, and a workflow based on your project description. Optionally creates a project state document to track phases, locked decisions, and session handoffs.
/ium:resume-project
Resume a project from a previous session — loads the workflow, reads the project state document, surfaces locked decisions, and briefs you on exactly where work stopped so you can continue without losing context.
/ium:help
Display all available /ium: commands with descriptions.
Workspace
Command
Description
/ium:workspace
Switch the active Contextium workspace with full context handoff — saves a note of what you were working on before leaving, then checks for a previous handoff note when arriving at the new workspace.
/ium:sync
Refresh the local Contextium cache from the current workspace so Claude has access to the latest files.
/ium:status
Show a health dashboard — auth status, active workspace, cache state, and available libraries, agents, and skills.
Content
Command
Description
/ium:new
Create a new file in a Contextium context library with guided prompts. Always writes complete content — never a placeholder.
/ium:create
Interactively create an agent, skill, or workflow in the current workspace.
/ium:search
Search across workspace documentation by query, filename, or tag. Respects loaded workflow scope — asks before expanding to the full workspace.
/ium:workflow
Load a Contextium workflow and sync its libraries, agents, and skills into the session context for immediate use.
Marketplace
Command
Description
/ium:marketplace
Browse, install, or publish skills, agents, and MCP servers on the Contextium Marketplace. Describe what you need in natural language — it finds the best matches and checks what you already have before suggesting installs.
Maintenance
Command
Description
/ium:update
Check for updates to @contextium/cli and @contextium/mcp-server, show installed vs latest versions, and update globally if confirmed.
Project state tracking
When you use /ium:new-project with the project planner option, Contextium creates a project-state.md document in your context library. This document tracks:
Current position — which phase you're on and what the next action is
Phases — each phase with its status (Not Started, In Progress, Complete)
Already built — capabilities confirmed by a codebase scan, so they don't get replanned
Decisions log — locked architectural and technical decisions with rationale, so sessions don't re-debate settled choices
Seeds — ideas to revisit at a future milestone, with trigger conditions
All project state is stored in Contextium — not locally — so it's available from any machine, survives context resets, and can be shared with your team. Running /ium:resume-project in a new session loads this document and restores full context.