Enigma - everything you need to work with a coding agent, in one command

What you get

Production engineering, deployed into your agent

Authored once, deployed to every agent you select - and kept current automatically.

Dynamic Skills, everywhere

Senior-engineering policies (security, testing, git, style, debugging...) on Claude Code, Codex, opencode and Kimi Code - each adapts to the agent and your config, loads by context, and updates itself.

Per-agent scoping

Keep a skill for Claude Code but not opencode. Turn any skill on or off per agent, from the CLI, the TUI or the dashboard.

Git security hooks

A portable, dependency-free commit guard for any repo: blocks secrets, .env files and dependency dirs before they are committed. Set it up once and the whole team inherits it.

Local dashboard

A loopback browser control panel for all of enigma - accounts, skills, settings, system cleanup - that also shows real Claude usage and measured savings. Nothing leaves your machine.

Multi-account + profiles

Several logins per tool without logging out, each in its own config dir, switched OS-agnostically. Profiles pin one account per tool and drive every launch.

Context compression

A native, dependency-free engine that shrinks large tool outputs, logs and text before they reach the model - the same information in far fewer tokens, reversibly.

Convention guardrails

Skills only persuade; guardrails enforce. A post-edit hook scans what the agent writes (UUID keys, Prisma as default ORM, your own rules) and feeds any violation back so it self-corrects - zero tokens until there is one. How guardrails work →

Verified completion

Agents report work as finished when it is not. On by default, enigma checks that claim against what the turn actually produced - unfinished markers, unimplemented paths, your own test command - and denies the stop when the evidence contradicts it. Zero tokens on a turn that claims nothing. How the completion gate works →

Local agent API

enigma api serves your local coding agents over one OpenAI-compatible HTTP API - Claude Code, and Codex/OpenCode where installed - all their tools, skills, MCP and sessions. One server, many backends: route per request by the model field. Loopback-only. How the API works →

Auto-detect skills by stack

enigma autoskills scans your project (packages, config files, gems) and installs the matching community skills for your stack - React, Next.js, Astro, Prisma, Rails and ~90 more - hash-verified and kept separate from the policy skills.

Marketplace packs

Add optional harness packs on demand - like Helio for bug bounty. Each runs in its own isolated agent context, so its skills and commands never load into or overload your normal agent. How packs work →

SSH connections

Save each server once - key or encrypted password, jump host, port forwards - and reach it with enigma ssh server1. Tunnel a remote port with a friendly 9090:db:5432 spec. Managed from the CLI, TUI and dashboard. How SSH works →

Built-in `/improve`

A slash command on every agent: an implement mode that edits a focused area (ui, security, performance...) and a read-only advisor mode that audits and writes plans for another agent to execute.

Built-in `/qa`

Where /improve reviews the code, /qa reviews the experience: it uses the product the way someone who depends on it would and fixes what fails them - dead ends, empty and error states, keyboard and screen-reader access, real-data extremes. How /qa works →

AI quality gate

An opt-in local gate that validates a branch in a disposable worktree - review, test, docs, lint, push, PR, CI - and opens a clean PR only when every check is green. Your checkout stays untouched. How the gate works →

Before / after

Same agent. Sharper output.

Enigma doesn't swap your model - it changes what comes out of it. The left is a coding agent on its defaults; the right is the same agent with enigma's policies, guard and output discipline.

Commit messages

git logWithout

update stuff

🤖 Generated with Claude Code
Co-Authored-By: Claude <noreply@anthropic.com>

git logWith Enigma

✨ feat(auth): add PKCE token refresh
🐛 fix(api):   retry 429s with backoff
♻️ refactor(db): extract query builder

Code structure

EnvManager.tsWithout

16 lines - singleton + cache

// EnvManager.ts
export class EnvManager {
  private static instance: EnvManager;
  private cache = new Map<string, string>();
  static getInstance() {
    if (!this.instance) this.instance = new EnvManager();
    return this.instance;
  }
  get(key: string, fb?: string) {
    if (this.cache.has(key)) return this.cache.get(key)!;
    const v = process.env[key] ?? fb;
    if (v == undefined) throw new Error("Missing " + key);
    this.cache.set(key, v); return v;
  }
}

env.tsWith Enigma

5 lines - same job

export function env(key: string, fb?: string) {
  const v = process.env[key] ?? fb;
  if (v == null) throw new Error("Missing " + key);
  return v;
}

Minimal-code (anti-overengineering): stdlib and native first, no speculative abstraction, the shortest diff that works.

Secrets & .env guard

bashWithout

$ git add .env
$ git commit -m "wip"
[main 4f2a1c8] wip
 2 files changed, 18 insertions(+)
$ git push        # .env + key now public

bashWith Enigma

$ git add .env
$ git commit -m "wip"
enigma-guard: BLOCKED - 2 problem(s):
  x .env  [env file with secrets]
  x src/config.ts:12  [secret: Anthropic API key]
commit aborted.

The commit guard ships in every repo you run enigma security in - it blocks secrets, .env files and dependency dirs before they ever reach a commit.

Descriptions & copy

README.mdWithout

Our revolutionary, next-generation platform — engineered from the ground up — leverages cutting-edge AI to seamlessly supercharge your workflow, unlock unprecedented synergy, and deliver world-class results — and so much more.

README.mdWith Enigma

Installs shared engineering skills into your coding agent, blocks secrets before they reach a commit, and manages everything from one local dashboard.

Token-efficient output

assistantWithout

Sure! I'd be happy to help you with that. Let me start by taking a careful look at the existing code so I can fully understand how everything is currently structured, and then I'll go ahead and make the changes needed to address the issue you've described.

assistantWith Enigma

Reading the existing code, then fixing the issue.

Three compression levels (lite / full / ultra) shape every reply: same technical content, answers sized to the question, fewer tokens per turn.

Less friction, fewer tokens

Defaults that stop wasting your time

Enigma turns off the noise and turns on the shortcuts the moment you install - so every session is faster, quieter and cheaper.

No approval prompts

Permission bypass is on by default, so the agent stops asking before every action and just works. A deliberate security trade-off you can opt out of globally or per agent.

No feedback surveys

Claude Code's feedback survey is switched off at install, so it never interrupts you mid-flow.

Right the first time

Specialized policy skills (security, testing, git, style, debugging...) mean fewer wrong turns and re-prompts - which means fewer tokens and less of your time.

No telemetry

GitHub CLI telemetry is disabled at install (and it sidesteps a Windows console-flash bug too). Nothing phones home.

Fewer tokens per turn

Opt-in token-efficient output and a native context-compression engine shrink prompts and tool output - the same work for fewer tokens.

Clean git history

The agent is kept from attributing commits to itself, so your history stays yours - nothing to undo later.

Headless components

Behaviour you install, not code you rewrite

Interaction logic is behaviour, and behaviour that has already been measured and tested belongs in a package. Add it as a dependency, or copy the source in and own it. No styles of their own - the look stays yours. How components work →

marquee

primitives

A looping row that can be grabbed and thrown - logo walls, testimonial strips, pill rails. Takes a speed in px/s, never a duration, so it does not accelerate every time you add an item.

Documentation →

cache

utils

Short-TTL read cache whose point is the in-flight deduplication: two components asking for one key make one request, not five, so a rate limit sees one.

Documentation →

notifications

utils

Ordering, dedupe by key and sticky errors, with dismiss timers that hold rather than run while the tab is hidden. Rendering stays yours.

Documentation →

Browse the components See the marquee API

Setup task	With Enigma	By hand
Shared skills across Claude Code, Codex, opencode, Kimi Code	One command	Per-agent, by hand
Skills kept current	Auto-sync + GitHub updates	Manual re-copy
Per-agent skill selection	Built in	Not available
Commit secret/.env guard	Portable, team-shareable	Ad-hoc scripts
Dashboard to manage it all	Local, loopback-only	None
Multi-account switching	OS-agnostic, no aliases	Shell hacks

Ship better code with your coding agent

Production engineering, deployed into your agent

Dynamic Skills, everywhere

Per-agent scoping

Git security hooks

Local dashboard

Multi-account + profiles

Context compression

Convention guardrails

Verified completion

Local agent API

Auto-detect skills by stack

Marketplace packs

SSH connections

Built-in `/improve`

Built-in `/qa`

AI quality gate

Same agent. Sharper output.

Defaults that stop wasting your time

No approval prompts

No feedback surveys

Right the first time

No telemetry

Fewer tokens per turn

Clean git history

Skills that adapt, not just deploy

Behaviour you install, not code you rewrite

marquee

cache

notifications

Manage enigma from your browser

One install instead of a dozen

Propose an improvement

Open a pull request

File an issue

Chat on Discord

Install Enigma

Production engineering, deployed into your agent

Dynamic Skills, everywhere

Per-agent scoping

Git security hooks

Local dashboard

Multi-account + profiles

Context compression

Convention guardrails

Verified completion

Local agent API

Auto-detect skills by stack

Marketplace packs

SSH connections

Built-in /improve

Built-in /qa

AI quality gate

Same agent. Sharper output.

Defaults that stop wasting your time

No approval prompts

No feedback surveys

Right the first time

No telemetry

Fewer tokens per turn

Clean git history

Skills that adapt, not just deploy

Behaviour you install, not code you rewrite

marquee

cache

notifications

Manage enigma from your browser

One install instead of a dozen

Propose an improvement

Open a pull request

File an issue

Chat on Discord

Install Enigma

Built-in `/improve`

Built-in `/qa`