Craft Easy Agentic¶
An autonomous task runner and developer dashboard for the Craft Easy framework.
Craft Easy Agentic runs Claude AI agents against your project's backlog — in parallel, in dependency order, with a live web dashboard for monitoring and control. It is the framework's development tooling: you write task specs, the runner implements them, and the dashboard shows you what's happening across every project you have registered.
What it does¶
- Runs Claude CLI agents autonomously — spawns parallel
claudeprocesses, each implementing one backlog task, respecting the dependency graph. - Web dashboard — real-time view of active tasks, progress, event log, KPIs, and cost. Responsive; works on mobile.
- Embedded Claude chat — Opus-pinned conversation panel for task creation, architecture review, and spec writing. Multi-modal (paste/drag/camera image input). Sessions persist across devices.
- Read-only file browser — browse any file in the project from the dashboard. Markdown, syntax-highlighted code, images, PDF — no write access by design.
- Git panel — multi-repo status (main + submodules + nested), Pull (
--ff-only) and Push from the UI; commit/merge/rebase stay in chat. - Drag-to-reorder priority queue — reshuffle the next-up task order on desktop, mobile, or keyboard, persisted as a transient
tasks/.priority-order.ymloverlay. - Multi-account rotation — wire in two or more Claude subscriptions and the runner rotates transparently on rate-limit, with chat-session continuity across accounts.
- Tools & Auth panel — diagnoses
gh/az/kubectl/git pushvisibility from the systemd service and one-click-enables host integration. - Memory-aware concurrency — cgroup-watched throttle that warns at 70 % and blocks Start at 85 % of
MemoryMaxto prevent OOM-kills. - Multi-project overview — one running instance serves every registered Craft Easy project from a single URL.
- Framework hub — sync canonical runner scripts, task templates, and gate configuration into every observed project.
Components¶
| Component | Purpose |
|---|---|
| Web dashboard | React + TypeScript UI, real-time via WebSocket |
| Rust server | Axum HTTP/WS, file watcher, state reconciliation |
| Bash runner | runner/run-backlog.sh — the actual Claude agent launcher |
| CLI | --headless for TUI, --once for CI snapshots, project subcommands |
Quick navigation¶
- New installation? Start with Installation.
- First time using the dashboard? See Web Dashboard.
- Managing multiple projects? See Multi-Project Management.
- Understanding Start/Stop/Pause/Force? See Runner Lifecycle.
- Using the Claude chat panel? See Embedded Chat.
- Setting up model routing and framework sync? See Framework Hub.
- Writing task specs? See Task Specification.
- Configuring acceptance gates? See Runner Config.
- Reordering the queue? See Priority Queue.
- Wiring in a second Claude account? See Multi-Account Rotation.
gh/az/kubectlfailing in chat? See Tools & Auth Panel.- Pull / push without leaving the dashboard? See Git Panel.
- SSH / CI / Termux usage? See CLI & Headless Mode.
Requirements¶
| Requirement | Version |
|---|---|
| Linux x86_64 | Any modern distribution |
| Claude CLI | Latest (npm install -g @anthropic-ai/claude-code) |
| Anthropic API access | Claude Max plan or API key |
| Node.js | 20+ (build only; not required at runtime) |
The binary is self-contained — it embeds the compiled frontend and has no external runtime dependencies (no Node.js, no Python required to run the server).
Relationship to the framework¶
Craft Easy Agentic is developer tooling — it is never deployed as part of your product. It observes any project that follows the craft-easy framework conventions and provides the task-runner and dashboard. The runner has no write access to your source code beyond what the Claude agents it spawns can do.
The canonical runner script (runner/run-backlog.sh) and task templates live here and are synced into observed projects via project sync.