ZERG Wiki
Parallel Claude Code execution system. Overwhelm features with coordinated zergling instances.
ZERG coordinates multiple Claude Code sessions to build features in parallel. You describe what to build, ZERG breaks the work into atomic tasks with exclusive file ownership, and multiple workers execute those tasks simultaneously -- organized by dependency levels so nothing conflicts.
Navigation
| Section |
Description |
| [[Getting Started]] |
Core concepts: levels, file ownership, spec-as-memory, verification |
| [[Installation]] |
pip install, requirements, Claude Code setup, Docker configuration |
| [[Quick Start]] |
Step-by-step first run through the full ZERG workflow |
| [[Your First Feature]] |
Guided walkthrough building a real feature with ZERG |
Reference
| Section |
Description |
| [[Command Reference]] |
All /zerg:* slash commands with flags and examples |
| [[Architecture Overview]] |
System internals: orchestrator, launcher, task graph, state management |
| [[Configuration]] |
.zerg/config.yaml options: workers, quality gates, resources, plugins |
| [[Tuning Guide]] |
Performance tuning: worker count, timeouts, resource limits |
| [[Troubleshooting]] |
Common issues, error messages, and recovery procedures |
| [[Debug Guide]] |
Using /zerg:debug, reading logs, inspecting task state |
| [[Contributing]] |
Development setup, coding standards, testing, PR workflow |
| [[Testing]] |
Test organization, running tests, coverage targets, writing new tests |
Cross-Cutting Capabilities
| Capability |
CLI Flags |
Description |
| Analysis Depth |
--quick/--think/--think-hard/--ultrathink |
5-tier depth control for worker analysis |
| Token Efficiency |
--no-compact (ON by default) |
Automatic compact mode with symbol/abbreviation system |
| Behavioral Modes |
--mode |
5 modes (precision, speed, exploration, refactor, debug) with auto-detection |
| MCP Auto-Routing |
--mcp/--no-mcp |
Capability-based MCP server selection with cost optimization |
| Engineering Rules |
via config |
25 YAML rules injected into worker context by file type |
| Improvement Loops |
--no-loop/--iterations (ON by default) |
Convergence-based iterative improvement cycles |
| Verification Gates |
via config |
Artifact storage, staleness detection, pipeline orchestration |
| TDD Enforcement |
--tdd |
Red-green-refactor protocol with anti-pattern detection |
Plugins and Context Engineering
| Section |
Description |
| [[Plugin System]] |
Overview of quality gates, lifecycle hooks, and launcher plugins |
| [[Plugin API Reference]] |
Abstract base classes, dataclasses, and registry methods |
| [[Context Engineering]] |
Command splitting, task-scoped context, security rule filtering |
| [[Context Engineering Internals]] |
Implementation details: splitting algorithm, token budgets, fallback behavior |
How ZERG Works
The workflow has four stages. Each stage requires explicit user approval before proceeding to the next.
Brainstorm Plan --> Design --> Rush --> Merge
(optional) | | | |
| v v v v
v Capture Generate Launch Merge
Discover requirements task parallel branches,
features, and specs graph workers run quality
create with file gates
issues ownership
Plan. You describe what to build. ZERG captures requirements through structured questions and writes them to a spec file.
Design. ZERG generates a technical architecture and breaks the work into a task graph. Each task owns specific files -- no two tasks touch the same file.
Rush. Multiple Claude Code instances (workers) execute tasks in parallel. Workers within the same dependency level run simultaneously. After all workers at a level finish, ZERG merges their branches before the next level begins.
Merge. The orchestrator merges worker branches after each level, runs quality gates (lint, typecheck, tests), and advances to the next level.
Key Commands
/zerg:init Initialize ZERG for a project
/zerg:brainstorm Discover features and create issues (optional)
/zerg:plan <feature> Capture requirements for a feature
/zerg:design Generate architecture and task graph
/zerg:rush --workers=5 Launch parallel workers
/zerg:status Check execution progress
/zerg:merge Manually trigger a level merge
/zerg:stop Stop all workers
/zerg:retry <task-id> Retry a failed task
See the Command Reference page for the full list of 26 commands.
Quick Links
- First time? Start with [[Installation]] then [[Quick Start]].
- Understand the concepts first? Read [[Getting Started]].
- Ready to build something? Jump to [[Your First Feature]].
1---2name: zerg-wiki3description: ZERG coordinates multiple Claude Code sessions to build features in parallel. You describe what to build, ZERG breaks the work into atomic tasks with exclusive file ownership, and multiple workers execute those tasks…4---5# ZERG Wiki67**Parallel Claude Code execution system.** Overwhelm features with coordinated zergling instances.89ZERG coordinates multiple Claude Code sessions to build features in parallel. You describe what to build, ZERG breaks the work into atomic tasks with exclusive file ownership, and multiple workers execute those tasks simultaneously -- organized by dependency levels so nothing conflicts.1011---1213## Navigation1415| Section | Description |16|---------|-------------|17| [[Getting Started]] | Core concepts: levels, file ownership, spec-as-memory, verification |18| [[Installation]] | pip install, requirements, Claude Code setup, Docker configuration |19| [[Quick Start]] | Step-by-step first run through the full ZERG workflow |20| [[Your First Feature]] | Guided walkthrough building a real feature with ZERG |2122### Reference2324| Section | Description |25|---------|-------------|26| [[Command Reference]] | All `/zerg:*` slash commands with flags and examples |27| [[Architecture Overview]] | System internals: orchestrator, launcher, task graph, state management |28| [[Configuration]] | `.zerg/config.yaml` options: workers, quality gates, resources, plugins |29| [[Tuning Guide]] | Performance tuning: worker count, timeouts, resource limits |30| [[Troubleshooting]] | Common issues, error messages, and recovery procedures |31| [[Debug Guide]] | Using `/zerg:debug`, reading logs, inspecting task state |32| [[Contributing]] | Development setup, coding standards, testing, PR workflow |33| [[Testing]] | Test organization, running tests, coverage targets, writing new tests |3435### Cross-Cutting Capabilities3637| Capability | CLI Flags | Description |38|------------|-----------|-------------|39| Analysis Depth | `--quick/--think/--think-hard/--ultrathink` | 5-tier depth control for worker analysis |40| Token Efficiency | `--no-compact` (ON by default) | Automatic compact mode with symbol/abbreviation system |41| Behavioral Modes | `--mode` | 5 modes (precision, speed, exploration, refactor, debug) with auto-detection |42| MCP Auto-Routing | `--mcp/--no-mcp` | Capability-based MCP server selection with cost optimization |43| Engineering Rules | via config | 25 YAML rules injected into worker context by file type |44| Improvement Loops | `--no-loop/--iterations` (ON by default) | Convergence-based iterative improvement cycles |45| Verification Gates | via config | Artifact storage, staleness detection, pipeline orchestration |46| TDD Enforcement | `--tdd` | Red-green-refactor protocol with anti-pattern detection |4748### Plugins and Context Engineering4950| Section | Description |51|---------|-------------|52| [[Plugin System]] | Overview of quality gates, lifecycle hooks, and launcher plugins |53| [[Plugin API Reference]] | Abstract base classes, dataclasses, and registry methods |54| [[Context Engineering]] | Command splitting, task-scoped context, security rule filtering |55| [[Context Engineering Internals]] | Implementation details: splitting algorithm, token budgets, fallback behavior |5657---5859## How ZERG Works6061The workflow has four stages. Each stage requires explicit user approval before proceeding to the next.6263```64Brainstorm Plan --> Design --> Rush --> Merge65(optional) | | | |66 | v v v v67 v Capture Generate Launch Merge68 Discover requirements task parallel branches,69 features, and specs graph workers run quality70 create with file gates71 issues ownership72```7374**Plan.** You describe what to build. ZERG captures requirements through structured questions and writes them to a spec file.7576**Design.** ZERG generates a technical architecture and breaks the work into a task graph. Each task owns specific files -- no two tasks touch the same file.7778**Rush.** Multiple Claude Code instances (workers) execute tasks in parallel. Workers within the same dependency level run simultaneously. After all workers at a level finish, ZERG merges their branches before the next level begins.7980**Merge.** The orchestrator merges worker branches after each level, runs quality gates (lint, typecheck, tests), and advances to the next level.8182---8384## Key Commands8586```87/zerg:init Initialize ZERG for a project88/zerg:brainstorm Discover features and create issues (optional)89/zerg:plan <feature> Capture requirements for a feature90/zerg:design Generate architecture and task graph91/zerg:rush --workers=5 Launch parallel workers92/zerg:status Check execution progress93/zerg:merge Manually trigger a level merge94/zerg:stop Stop all workers95/zerg:retry <task-id> Retry a failed task96```9798See the Command Reference page for the full list of 26 commands.99100---101102## Quick Links103104- **First time?** Start with [[Installation]] then [[Quick Start]].105- **Understand the concepts first?** Read [[Getting Started]].106- **Ready to build something?** Jump to [[Your First Feature]].
tools-only