An agent harness and development workflow for orchestrating AI-driven programming tasks.
Use the simple CLI ark to define your AI workflow and manage coding agents in your project.
See the deck for design details.
Status: Ark is experimental and some features are unstable. See the Future Plan.
Ark defines the next generation of AI-native software engineering work: structured workflows, agent collaboration, and tool orchestration.
Ark provides:
- A simple CLI for managing AI-native software projects.
- A workflow harness for coordinating Claude Code, Codex, and other coding agents.
- Sandbox for isolated execution environment.
- Session-level journals for tracking agent activity and workspace state.
- Task-level development records with integrated worktree management.
- Specification management for projects and features.
Prebuilt binaries are available for macOS, Linux, and Windows on every tagged release.
npm install -g @anekoique/arkcargo install --git https://github.com/Anekoique/ark ark-cli --lockedConfirm the installation:
ark --versionFrom the root of a project where you want to use Ark:
ark initOn first run, Ark scaffolds:
.ark/
├── workflow.md # the rules of the game
├── config.toml # worktree, workspace, upgrade, sandbox settings
├── templates/ # PRD, PLAN, REVIEW, VERIFY, SPEC
├── tasks/ # active + archived tasks
└── specs/
├── project/ # user-authored project conventions
└── features/ # feature specs promoted from deep-tier tasks
.claude/commands/ark/ # slash commands (Claude Code; Codex/OpenCode equivalents)
├── quick.md # /ark:quick
├── design.md # /ark:design [--deep]
├── research.md # /ark:research
└── … # commit, resume, discard, record, spec-extract, spec-audit
CLAUDE.md # managed block pointing the agent at .ark/
Open your agent in the project and start a task:
/ark:quick fix typo in readme
/ark:design add rate-limit middleware
/ark:design --deep refactor auth layer
/ark:research compare WAL strategies
See .ark/workflow.md for the full workflow.
Pick the tier that fits the different tasks.
| Tier | For |
|---|---|
| Quick | Reversible in one commit, no new abstractions |
| Standard | Feature work with testable scope |
| Deep | Architectural or new subsystem |
| Research | Knowledge-gathering; corpus is the deliverable |
ark provides the following commands for managing its presence in a project:
| Command | What it does |
|---|---|
ark init |
Driven project with ark |
ark load |
Restore from .ark.db if present. |
ark unload |
Snapshot .ark/, managed blocks, and the Ark hook into .ark.db. |
ark remove |
Wipe Ark fully. |
ark upgrade |
Refresh embedded templates to the current CLI version. |
ark context |
Get current project context which driven ark |
ark archive |
Stabilize some features. |
ark cleanup |
List or remove worktrees of closed tasks. |
ark sandbox |
Run agent inside a isolated enviroment. |
User-authored files inside owned dirs survive an unload → load round-trip losslessly.
ark sandbox provide isolated execution environment.
ark sandbox create # start the box
ark sandbox enter # launch the agent CLI inside (--shell for bash)
ark sandbox rm # tear it downConfigure the image, env passthrough, and host-config sharing under [sandbox] in .ark/config.toml. Requires docker on PATH.
Ark is highly inspired by and learns from:
MIT