Cadence

A stack-agnostic, self-improving AI development framework as a Claude Code plugin.

View the Project on GitHub nik190799/cadence

Quickstart

Five minutes from claude plugins add to the framework improving itself on your repo.

Prerequisites

1. Install the plugin

Inside any Claude Code session:

/plugin marketplace add nik190799/cadence
/plugin install cadence@cadence

The first command registers the Cadence repo as a plugin marketplace. The second installs the cadence plugin from it. You only need to do this once per workstation; the plugin is then available in every session.

Verify:

/plugin list

You should see cadence@cadence v0.1.0.

2. Initialize the framework in your project

/cadence-init

Cadence detects your project’s stack from common markers (package.json, pyproject.toml, go.mod, Cargo.toml, pubspec.yaml, build.gradle) and prefills .cadence/cadence.yaml with sensible default commands.

It writes:

Review .cadence/cadence.yaml and adjust as needed.

3. Verify the gates fire on your existing code

/cadence-verify

This runs the verify pipeline (format → lint → boundary check → test) using your project’s tooling and renders a Definition-of-Done aligned checklist. Existing issues become Cadence findings against named rules.

4. Launch a team for your next feature

/cadence-launch "favorites tab"

Cadence fills in TEAM_LAUNCH_TEMPLATE.md placeholders. You can choose to spawn four subagents (data engineer, feature engineer, tester, reviewer) who coordinate via a shared task list — or use the prompt yourself as a single session.

5. Run the retrospective

After the work is done:

/cadence-retro

The Reviewer walks the four-layer fix mapping: for each issue observed, is it auto-catchable (Layer 3), a missing pattern (Layer 1), a process gap (Layer 2), or a launch ambiguity (Layer 4)? Approved fixes land immediately as new ADRs, PATTERNS entries, boundary rules, or DoD lines. The change is recorded in FRAMEWORK_CHANGELOG.md.

Your framework just got smarter on YOUR codebase.

What next