Slash Commands
ContextKit 2.0 installs five small skills: /ck, /doc, and /learn, plus two add-ons, /goal and /headline. Each skill is a thin wrapper around a workflow file in .contextkit/commands/.
Overview
Skills are generated from one registry and rendered for each tool, so they behave the same everywhere. In Claude Code and Cursor you run them as /name. In Codex you run them as $name.
How it works
1. ck claude writes skills to .claude/skills/<name>/SKILL.md
2. ck codex writes .codex/skills/ and the portable .agents/skills/
3. ck cursor writes .cursor/prompts/<name>.md and .agents/skills/
4. Every skill reads its workflow from .contextkit/commands/
The Five Skills
| Skill | Usage | What it does |
|---|---|---|
| /ck | /ck, /ck setup, /ck update, /ck tidy | Status, first-time setup, finishing an upgrade, and keeping context lean |
| /doc | /doc, /doc arch, /doc feature [name], /doc component <path> | Update docs and standards affected by your change, or write a doc at a given level |
| /learn | /learn "rule", /learn | Turn a correction into one rule in the right standards file |
| /goaladd-on | /goal "idea", /goal, /goal status, /goal done | Run a long task against a one-page goal file. See /goal → |
| /headlineadd-on | /headline, /headline off, /headline pause | Short, result-first replies, always on. See /headline → |
Upgrading from 1.x? The older skills were removed or folded into these five. The upgrade guide lists each one and its replacement.
/ck
Manages the project's AI context from inside your AI tool.
/ck
Status: what's set up and what needs attention.
/ck setup
First run. Analyzes the codebase and writes the standards, the always-loaded .contextkit/standards/index.md, and the context_rules in .contextkit/config.yml that decide which topic file loads for which paths.
/ck update
Finishes an upgrade after ck update. Works through .contextkit/pending-cleanup.md and asks before each change. For custom content in CLAUDE.md you choose: Move to AGENTS.md, Keep in CLAUDE.md, or Delete. For unfinished work from removed 1.x workflows you choose: convert to a goal, archive, or keep.
/ck tidy
Shrinks always-loaded context, prunes generic or linter-enforced rules, wires guides that nothing loads into context_rules, and promotes corrections that keep repeating into standards.
/doc
Keeps docs and standards in step with the code.
- No argument: finds docs and standards that reference the code changed on this branch and updates the stale ones. On a feature branch it lists the updates first. For an npm package it also suggests a version bump and checks
CHANGELOG.md. /doc arch: architecture doc (Level 1)/doc feature [name]: feature or page README (Level 2)/doc component <path>: component README, colocated (Level 3)
/learn
Turns a correction into one rule so the mistake doesn't repeat. It checks for an existing rule first, writes to the right file in .contextkit/standards/ (topic file, index.md, glossary, or architecture), and logs it with ck note.
/learn "we use the MUI wrapper, never raw MUI"/learnClaude Code
Skills in .claude/skills/
Setup
Usage
/ck setup/doc/goal "add CSV export to reports"Files created
.claude/skills/ck/SKILL.md.claude/skills/doc/SKILL.md.claude/skills/learn/SKILL.md.claude/skills/goal/SKILL.md.claude/skills/headline/SKILL.md
Cursor
Prompts in .cursor/prompts/
Setup
Files created
.cursor/prompts/ck.md,doc.md,learn.md,goal.md,headline.md.agents/skills/<name>/SKILL.md(portable copy)
Codex
Skills in .codex/skills/ and .agents/skills/
Setup
Usage
$ck setup$goalGlobal Skills & Plugin
The add-ons don't need ContextKit in the project, so you can share them on their own. Global install writes to ~/.claude/skills/:
ck install --global goal ck install --global headline ck update --global # regenerate; also removes retired pre-2.0 global skills ck uninstall --global <name> # remove one skill, including retired pre-2.0 ones
ContextKit is also a Claude Code plugin. It ships /ck, /doc, /learn, /goal, and /headline:
claude plugin marketplace add nolrm/contextkit claude plugin install contextkit@contextkit-marketplace
Was this helpful?
Help us improve the documentation by sharing your feedback.