Multi-agent software development orchestration system with built-in quality assurance, skill evolution, and memory consolidation.

SwarmX coordinates specialized AI agents to collaborate on software development tasks. It provides:

• Agent Orchestration: Centralized task coordination via Conductor agent
• Skill System: Event-triggered skill invocation with evolution
• Quality Assurance: File ownership enforcement and code review workflows
• Intelligence: Episode logging and skill improvement tracking

┌─────────────────────────────────────────────────────────────┐
│                      Orchestration Pool                      │
│   Conductor  │  Spec  │  Architect  │  Reflection  │  ...   │
└──────────────────────────────┬──────────────────────────────┘
                               │ assigns tasks
                               ▼
┌─────────────────────────────────────────────────────────────┐
│                       Execution Pool                         │
│      Backend     │     Frontend     │      DevOps      │    │
└──────────────────────────────┬──────────────────────────────┘
                               │ results
                               ▼
┌─────────────────────────────────────────────────────────────┐
│                       Assurance Pool                        │
│         Arbiter         │       Skill Evolver        │     │
└─────────────────────────────────────────────────────────────┘

# Build the project
cargo build --workspace

# Run tests
cargo test --workspace

# Start the API server
cargo run --bin swxd

• Architecture
• Specification
• Data Design
• API Contracts
• Shared Types

Agents are specialized AI workers assigned to specific tasks:

• Orchestration Pool: Conductor, Spec, Architect, Reflection, etc.
• Execution Pool: Backend, Frontend, DevOps, Data, Security
• Assurance Pool: Arbiter, Skill Evolver, etc.

Tasks are work items with:

• Priority (Critical, High, Normal, Low)
• State (Pending → Completed)
• Parent/child relationships (epics)
• Context tags for categorization

Skills are reusable workflows triggered by events:

The Intel system evolves skills based on usage:

1. Log successful task completions as episodes
2. Propose skill improvements
3. Apply changes with appropriate safety tier

1. Add AgentId variant in crates/swx-core/src/agent.rs
2. Assign to appropriate AgentPool
3. Implement message handling in swx-actors

1. Create skill definition in crates/swx-core/src/skill.rs
2. Add trigger mapping
3. Implement skill scripts

cargo new crates/swx-new-crate

Add to workspace in Cargo.toml.

MIT