Stop re-explaining
your project to AI.

Contextkit is the missing workflow layer for Claude Code, Codex and every AI coding agent — persistent project memory, guided planning, a built-in slop detector, and short commands you actually remember.

See how it works
Ask-first ruleClarity Engine37-tell detectorSlop guardrails17 skills
Claude CodeCodexAntigravityGemini CLICursorWindsurfZedAGENTS.md
Claude CodeCodexAntigravityGemini CLICursorWindsurfZedAGENTS.md
Before → After

From scattered prompts to a
streamlined coding workflow

Fewer misunderstandings, less re-explaining, and every agent session already knows your project.

Long promptsRe-explainingRandom commandsLost sessionsBlank terminalGuessing
Contextkit
Your coding co-pilot
Auto-restored context
Clarified requirements
Slop detection
Short commands
Contextual lessons
Our signature feature

Meet the
Slop Detector.

After every AI change, Contextkit scans the diff for the telltale signs of AI-generated slop — repetitive scaffolding, hallucinated APIs, throwaway abstractions, cosmetic edits, and obvious pattern-copy. Then it notifies, warns, or blocks the commit.

  • Repetitive patterns & duplicated components
  • Suspicious styling, dead branches, unused props
  • Hallucinated imports and non-existent APIs
  • Cosmetic-only refactors that add complexity
3.2×
fewer bad diffs
-58%
revert commits
Slop scan · 42 files
3 issues
!
utils/format.ts
Duplicated of lib/format.ts · 92% match
components/Hero.tsx
Suspicious styling — inline hex colors
api/users.ts
Import from 'next-auth/react' — package not installed
routes/index.tsx
Clean
styles.css
Clean
Enforce corrections before commit
What's inside

A few skills.
Each one, deep.

Not 159 commands you'll never remember. 17 skills, one command each, one install. Drop it into Claude Code or Codex and the whole engine — skills, detector, hooks — runs natively.

(01)
always on

The core

Ask before you type. Be direct. This is the rule the entire kit stands on — plus the setup wizard that installs it everywhere.

/start/tour— the AGENTS.md rule— session greeting
(02)

Build cleanly

One move at a time. Prove it runs. Trim the code that shouldn't have shipped. No more forty-file diffs for a two-line ask.

/step/check/fix/review/lean
(03)

Ship & survive

Get it out the door, catch the secrets it almost leaked, and roll back cleanly when a turn goes sideways.

/ship/secrets/recover/explain
(04)
37-tell detector

Doesn't look AI-made

A real scanner for the 37 giveaways that scream 'AI generated this' — plus the skills to quietly fix each one.

/design-intake/design-check/polish
(05)

Reach further

Live docs and web search through crawl4ai and exa, plus the Claude Code and Codex tricks nobody told you about.

/connect-mcp/claude-tips/codex-tips
(06)
opt-in

Voices

Flip into a teaching tone when you're learning, a buddy tone when you're tired. Direct stays the default.

— teaching— buddy

17 skills. One that matters most: it asks first.

Works anywhere AGENTS.md is read — Claude Code, Codex, Antigravity, and the ones landing next month.

Fewer prompts, more shipping

Each skill is one
short command.

17 skills, 17 commands — each one covers a full workflow. No prompt engineering needed. Contextkit handles what happens behind each slash; you just remember the word.

claude-code · contextkit — 8 of 17 skills
/startAsk first, plan the project, write CLAUDE.md
/stepOne step at a time, checkpoint when it works
/checkProve it runs — show real evidence, not vibes
/fixRoot cause only. 3 strikes then rethink
/reviewDiff against the goal before you accept
/secretsCatch leaked keys before anything goes public
/polishFix slop tells by root cause, not decoration
/shipDeploy, set prod env vars, verify the live URL
Type any command, or press ? for help
One-time price, forever yours

Cheaper than one hour of
cleaning up the slop.

No subscription. No per-seat. No renewal. Pay once, own it forever — including every future update.

Best-sellerThe kit
$5
$29once

Everything you need to make your agent ask first and build clean.

  • All 17 skills plus the 37-tell slop detector
  • Auto hooks: anti-slop, safety, and the greeting
  • Teaching and buddy voices, direct by default
  • Lifetime updates as Claude Code and Codex move
Get the kit

Secure checkout via Dodo Payments. Instant download after purchase.

Claude Code Codex Antigravity any AGENTS.md reader

Questions, answered

Is Contextkit another AI coding assistant?+

No. Contextkit does not generate code. It sits between you and your existing AI coding agent (Claude Code, Codex, etc.) and upgrades it with memory, planning, teaching and guardrails.

Which agents are supported?+

Claude Code, Codex and Antigravity today. Gemini CLI, Cursor, Windsurf and Zed are next. Anything that can read an AGENTS.md-style project file will eventually work.

What exactly does the slop detector look at?+

After every AI change it scans the diff for repetitive patterns, duplicated files, hallucinated imports, cosmetic-only refactors and suspicious styling. You choose whether it notifies, warns or blocks.

Can I skip the clarification questions?+

Yes. Type “just do it” or set the Clarity Engine to Relaxed or Disabled. Strict is recommended when the request touches many files.

What happens to my project data?+

Roadmap, milestones and decisions live in your project folder as plain files. Nothing gets locked into Contextkit and everything is portable.

Ship real projects.
Not just prompts.

Join the early access and help shape the workflow layer AI coding has been missing.

Contextkit

The workflow layer for AI coding agents. Memory, planning, slop detection, and short commands that work across every agent you use.

Product
Resources
Company
Contextkit