Selects optimal subagent topology (hierarchical, mesh, ring, star) based on task structure and coordinates multi-agent swarm execution for complex, multi-file coding tasks.
npx claudepluginhub gadaalabs/claude-code-on-steroidsThis skill uses the workspace's default tool permissions.
**LEGION** — *A legion is a large, coordinated force organized into precise, independent units.*
Creates isolated Git worktrees for feature branches with prioritized directory selection, gitignore safety checks, auto project setup for Node/Python/Rust/Go, and baseline verification.
Executes implementation plans in current session by dispatching fresh subagents per independent task, with two-stage reviews: spec compliance then code quality.
Dispatches parallel agents to independently tackle 2+ tasks like separate test failures or subsystems without shared state or dependencies.
LEGION — A legion is a large, coordinated force organized into precise, independent units. When invoked: selects the optimal agent topology for the task — parallel (independent tasks), pipeline (sequential stages), hierarchical (orchestrator + specialists), or mesh (peer review) — and coordinates execution across all agents.
Core principle: Match agent topology to task structure — the right coordination pattern prevents chaos and maximizes parallelism.
Announce at start: "Running LEGION to select agent topology and coordinate."
Independent tasks, no coordination needed → STAR (see patterns/star.md)
Sequential stages, each transforms previous output → RING (see patterns/ring.md)
Exploration, multiple hypotheses, no clear structure → MESH (see patterns/mesh.md)
Plan execution, coordinator + specialists → HIERARCHICAL (see patterns/hierarchical.md)
Load the matching pattern file for full prompt templates and examples.
| Topology | Parallelism | Best For |
|---|---|---|
| Hierarchical | Moderate | Plan execution, structured tasks |
| Mesh | High | Exploration, research, bug investigation |
| Ring | Low (sequential) | Data pipelines, multi-stage transforms |
| Star | Very High | Independent parallel work, batch processing |
| Task Complexity | Swarm Size | Topology |
|---|---|---|
| Single file edit | 1 agent | N/A (no swarm) |
| 2-5 files | 2-3 agents | Hierarchical |
| 5-10 files | 3-5 agents | Hierarchical or Star |
| Multi-module | 5-7 agents | Hierarchical-Mesh |
| System-wide | 7-15 agents | Adaptive with phases |
10 agents: coordination overhead exceeds parallelism benefit. >15 agents: decompose into separate projects.
Complex projects with both exploration and execution:
PHASE 1: Exploration (Mesh) — explore angles, synthesize requirements
PHASE 2: Planning (Hierarchical) — coordinator writes plan, decomposes tasks
PHASE 3: Execution (Hierarchical) — workers execute, coordinator reviews
PHASE 4: Integration (Mesh) — collective debugging and integration
Majority: 3 agents agree A, 2 agree B → proceed with A, document dissent. Weighted: Security decision → security-architect vote 3x weight. Byzantine: f < n/3 faulty agents. With 4 agents: tolerate 1 faulty opinion. If no majority → escalate to human.
Auto-trigger when ALL of:
Auto-trigger announcement:
DETECTING: This task touches N independent files with no cross-dependencies.
AUTO-TRIGGERING: <topology> swarm (N spokes/workers)
Topology rationale: [reason]
Do NOT auto-trigger for: bug fixes (investigate first), tasks < 3 files, shared state changes.
| Skill | Integration |
|---|---|
phantom | Hierarchical topology by default |
architect | Mesh topology for exploration phase |
hunter | Mesh topology for multi-angle investigation |
blueprint | Plan should specify recommended topology |
tribunal | Star topology for parallel file review |
Never:
Topology determines success — match structure to task
Hierarchical for execution | Mesh for exploration
Ring for pipelines | Star for parallel work