Features
Gem Team combines four pillars: Quality by default, Smart and efficient, Works with your tools, and Learns and improves.
Ship With Confidence
Core acceptance checks run automatically; specialist audits are routed by task requirements and changed-scope risk:
- Library-first - prefer maintained libraries over custom implementations.
- Plan validation - validate MEDIUM/HIGH plans for ordering, ownership, correctness, and contract integrity.
- Risk-based review - use standard, high, or critic review for high-risk signals, explicit requests, or insufficient evidence.
- Security and compliance - review changed code, configuration, and integrations; route security-sensitive work to specialists.
- Accessibility - configurable WCAG 2.2 AA checks for contrast, ARIA, focus indicators, and touch targets (
none,basic,full). - Visual regression - compare screenshots against a configurable threshold.
- Simplification and clarity - apply targeted design simplification and answer with the action first.
Test Automatically
TDD and acceptance checks are part of the workflow:
- Browser E2E - flow assertions, visual evidence, and console/network capture.
- Mobile E2E - iOS and Android checks with Detox, Maestro, or Appium.
- Performance - cold-start TTI, memory, frame rate, and bundle-size checks.
- Platform behavior - safe areas, keyboards, permissions, dark mode, and haptics.
Match Model Strength to Task Complexity
Model routing assigns configured tiers to delegated agents:
| Tier | Best for |
|---|---|
| Premium | Planning, debugging, and review |
| Explore | Research, implementation, testing, documentation, and routine work |
Choose models and providers in .gem-team.yaml; the Orchestrator applies the tier when delegating work.
Design Consistently
Design systems and accessibility are built into the workflow:
- Design systems - palettes, type scales, spacing, shadows, and design movements in a nine-section
DESIGN.md. - Mobile - iOS HIG, Material 3, safe areas, Dynamic Island, and 44pt/48dp touch targets.
- Accessibility-first - 4.5:1 contrast, reduced motion, and semantic HTML/ARIA.
- Implementation prompts - design output includes agent-ready prompts.
Deploy Without Fear
Dedicated agents handle infrastructure and CI/CD:
- Infrastructure - Docker, Kubernetes, and AWS/GCP/Azure.
- CI/CD - PR -> staging -> smoke -> production with configurable approval gates.
- Health and recovery - endpoint verification, resource monitoring, and rollback strategies.
- Mobile release - EAS Build/Update, Fastlane, TestFlight, and phased Google Play rollouts.
- Idempotent operations - safe to run again.
Resumable by Design
Every MEDIUM/HIGH task gets a persistent plan ID and plan.yaml. Work can be paused, resumed, or extended without losing context:
- Exact plan IDs - no fuzzy matching or inference; continuation requires the exact ID.
- Ordered waves - each wave completes before the next begins; evidence carries forward through scoped handoffs.
- Parallel execution - independent tasks run in parallel within waves; overlapping ownership is serialized to prevent conflicts.
- Bounded replans - replans revise execution details, not the user's goal. A maximum of two replans prevents infinite loops.
- Relational invariant fallbacks - when an agent output violates a conditional requirement, the Orchestrator infers intent and fills the gap instead of rejecting valid work.
See Workflow -> for the full pipeline.
Cross-Cutting Capabilities
These run across all workflows. Each links to the page with the full detail:
| Capability | Covered in |
|---|---|
| Workflow engine: phases, routing, gates, and resumable plans | Workflow -> · Core Concepts -> |
| Specialist agents and Orchestrator | Agent Team -> |
| Task handoffs, memory tiers, and skill extraction | Core Concepts -> |
| Cost control: model routing, output hygiene, and cache math | Optimizations -> |
| Learning and reuse: persistence, batching, and checkpoints | Optimizations -> |