Everything you need to run AgenKit.

Overview

AgenKit is a guided AI engineering team you install into your coding tool. It ships a full roster of specialist agents, the process skills that guide them, and a single command, /agenkit, that runs the whole lifecycle: brainstorm → architecture → plan → test-first build → review → ship, pausing for your approval at each step.

It works across seven harnesses from one license key, installs in one command, and every artifact is SHA-256 verified before it touches your project. Once installed, it runs locally and keeps working offline.

Requirements

• Node.js 18+: the agenkit CLI runs with npx (nothing to install globally).
• A supported harness: Claude Code, Cursor, Codex, OpenCode, Antigravity, GitHub Copilot, or Kimi Code.
• A license key: issued at checkout; the CLI stores it locally. No account to manage.

Quickstart

Three steps take you from a license key to a working AI team in your project.

$ npx agenkit activate <license-key>

$ npx agenkit install engineering-kit

$ /agenkit

Prefer a global binary? npm install -g agenkit, then drop the npx prefix.

Supported harnesses

One kit, compiled to the exact file layout each tool expects. agenkit install auto-detects your harness; override with --harness <name>.

Kimi Code is plugin-based: after install, run /plugins install .agenkit/engineering-kit, start a fresh session, then /agenkit:engineering (type /agenkit to find it).

The /agenkit workflow

You describe the task once. The Conductor hands each phase to the right specialist and pauses for your approval, the way a senior team ships software. Nothing is implemented until the spec, architecture, and plan are each approved.

1. 0
   OrientContext Engineer

   Builds or refreshes local memory so every later phase can query the map instead of re-reading files. Fast, local, no checkpoint.

2. 1
   BrainstormConductor

   Turns your rough idea into a clear, written spec through focused questions. Saved to docs/specs/. You approve before anything else.

3. 2
   ArchitectureBackend Architect + specialists

   Designs folders, module boundaries, interfaces, data models, and the failure strategy (idempotency, retries, reconciliation).

4. 3
   PlanConductor

   Turns the spec + architecture into a bite-sized, test-first implementation plan with exact file paths. Saved to docs/plans/.

5. 4
   BuildThe right specialist per task

   Implements test-first (RED-GREEN-REFACTOR) in an isolated worktree, committing frequently.

6. 5
   ReviewCode Reviewer + Debugger

   Two-stage review: spec compliance, then code quality. Critical findings block progress until fixed.

7. 6
   ShipInfra/DevOps + Security + Perf

   Sets up CI/CD and deployment, runs a final security gate, and confirms metrics/SLOs. You choose merge / PR / keep.

Specialist agents

Eleven single-responsibility specialists. The Conductor coordinates them; each brings the production concerns a generalist tends to skip.

Local memory

Local memory is a deterministic, on-your-machine index of your codebase. Instead of re-reading and grepping files every session, the agents query the index for “where does X live / what connects to Y” and jump straight to the exactfile:line. Building it is local and LLM-free, so it costs ~0 tokens, and querying it costs a fraction of reading whole files.

You don't run anything: it's automatic

When you run /agenkit, a Phase 0 “Orient” step builds or refreshes memory for you, and an always-on rule nudges the agent to query memory before reading files (even outside /agenkit). The commands below are for when you want to drive it by hand.

What gets stored

Everything lives in a .agenkit/memory/ folder in your project:

The short digest and notes are committed so your team starts warm; the heavy raw index stays local and rebuilds on demand. It's a fast navigation aid, not a compiler. When unsure, the agent just opens the exact file it pointed to.

Memory commands

Optional: for driving memory manually.

$ npx agenkit memory build

$ npx agenkit memory query "where is the login endpoint?"

$ npx agenkit memory explain "UserService"

Updating

Your subscription includes every new release. Pull the latest kit into a project with:

$ npx agenkit update

This refreshes the agents, skills, command, and rules in place; your own files are never overwritten. If you installed the CLI globally, refresh it too:

$ npm install -g agenkit@latest

CLI reference

Add --help to any command for its full options.

Troubleshooting

“command not found” after install

Use npx agenkit ... (no global install needed), or run npm install -g agenkit and open a new terminal.

The kit didn't appear in my tool

Confirm the install reported your harness. If auto-detection missed it, re-run with the explicit flag, e.g. npx agenkit install engineering-kit --harness cursor. Then restart the tool so it re-scans.

“/agenkit” doesn't show up

Triggers differ per harness; see Supported harnesses. In GitHub Copilot it's @agenkit; in Codex you run the AgenKit skill by name; in Kimi it's /agenkit:engineering.

Memory answers seem out of date

Run npx agenkit memory status; if it reports drift, run npx agenkit memory update. The /agenkit workflow refreshes it automatically after the build phase.

My license isn't recognized

Re-run npx agenkit activate <key>. If it still fails, your subscription may have lapsed; manage it from your billing portal.

Support

Questions, bugs, or feature requests? Reach us on Telegram at @agenkit. For billing, see Pricing & FAQ.