Stella Loop documentation
Everything you need to run loop-based development with Stella Loop — from your first workspace to standing intents, tournaments, and a fleet of coding agents working through the CLI.
Getting started
What is Stella Loop An agent-native project management tool for loop-based development — what it is, who it serves, and how it differs from an issue tracker. Getting started From a fresh workspace to your first completed loop: connect a repository, set the North Star, fire an intent, promote, and ship.
Core concepts
The loop The full lifecycle: signals, North Star, intents, analysis, proposals, promotion, epics, plan, implement, review — and the edges that make it a loop. North Star The constellation: typed, weighted, versioned North Star documents that analyzers score against and every piece of work traces back to. Signals The loop's front door: frictionless capture, connector ingestion, triage dispositions, demand clusters, and resolution notices. Intents The statement of aim that starts every loop turn: directed and North Star-driven archetypes, standing triggers, and preserved run history. Analysis Analyzers, reports, and scores: the evidence engine — parallel runs, versioned rubrics, comparable 0–100 scores, and a dashboard that moves. Proposal pool Sibling sets, priority bands, dependencies, pruning — and promotion, the commitment point where epics are born. Epics and tasks The delivery container: stage pipeline with loop-backs, tournament candidates and branch-sets, and race-safe task claiming. Plan Specs and decomposition: pluggable spec tools (OpenSpec by default), approval as a checkpoint, and the task handoff. Review Panels with explicit lenses, findings, fix and re-spec verdicts, tournament selection, ordered merges, and the learnings that seed the next loop. The loop engine What makes the stages one loop: iterations, checkpoint gates, the autonomy dial, model tiers, and budgets with visible stops. Provenance The chain that answers why work exists: epic → proposal → report → intent → signal, with display IDs that make it legible everywhere.
Agents
Agents Agents are first-class users: enforced parity across CLI, API, and MCP, API keys, and the canonical self-discovering agent session. CLI The stella command line: self-discoverable help, stable JSON output, a strict exit-code contract, and the full command tree. API The public /api/v1 surface: API keys, one envelope, display-ID addressing, idempotency, rate limits, webhooks, and the event feed. MCP The generated Model Context Protocol server: stella mcp serve, tool naming, annotations, and configuration for MCP-speaking agents.
Platform
Web app The keyboard-first interface: command palette, global search with typed filters, realtime liveness, and the shortcuts worth learning. Desktop app The macOS app: global summon, tray presence, native actionable notifications, deep links, offline reading, and staged updates. GitHub integration The GitHub App, repository linking, branch-sets, change sets with computed approval, ordered merges, and webhooks-as-hints resilience. Signal connectors Automatic signal ingestion from GitHub Issues, Slack, Sentry, inbound email, and generic signed webhooks — with health you can see. Notifications One inbox, four channels: in-app, desktop, email and digests, and webhooks for agents — with inline approvals and honest unread counts.