| name | boomerang-init |
| description | Initialize and personalize Boomerang agents for your project. Run once at project start, and again anytime you want to refresh agents as the project evolves. |
Boomerang Init
Description
Initialize and personalize Boomerang agents for your project. Run once at project start, and again anytime you want to refresh agents as the project evolves.
Usage: /boomerang-init
Hard Rules for Agent Customization
PROTECTED: Core Prompting
The following elements are PROTECTED and must NEVER be modified, removed, or replaced by boomerang-init:
- The 6-Step Boomerang Protocol — Memory → Think → Delegate → Git Check → Quality Gates → Save Memory
- Agent selection rules — Which agent types route to which sub-agents
- Mandatory step ordering — super-memory query first, sequential thinking second
- Quality gate requirements — lint → typecheck → test enforcement
- Sub-agent requirements — super-memory and sequential-thinking mandates in prompts
PERMITTED: Append-Only Customizations
boomerang-init MAY only append to the following:
- Agent personas — Add project-specific domain context to the END of agent descriptions
- Tool preferences — Add project-specific tool recommendations
- Conventions — Add coding standards and patterns
- Success metrics — Add project-specific definitions of success
Forbidden Operations
boomerang-init must NEVER:
- Delete or replace agent system prompts
- Remove quality gate requirements
- Skip or reorder mandatory protocol steps
- Remove super-memory or sequential-thinking requirements
- Change agent routing logic
- Modify the core orchestrator instructions
Customization Format
When customizing an agent, use this append-only format:
## Project-Specific Context (Appended by boomerang-init)
### Domain
[Project domain description]
### Conventions
- [Convention 1]
- [Convention 2]
### Stakeholders
- [Who uses this project]
### Success Metrics
- [What good looks like]
This ensures all customizations are additive and can be easily identified and removed if needed.
What It Does
- Analyzes your project - Examines structure, language, README, package.json
- Identifies project type - Determines if special agents are needed
- Customizes base agents - Tailors prompts to your project context
- Creates project-specific agents - Adds specialists for your domain
- Updates AGENTS.md - Documents your custom agent team
Tiered Memory Protocol
This project uses a tiered memory architecture with two modes:
- Fast Reply (TIERED): Quick MiniLM search with BGE fallback for speed
- Archivist (PARALLEL): Dual-tier search with RRF fusion for maximum recall
When Saving:
- Routine work (quick analyses, Explorations): Use standard
super-memory_add_memory
- High-value work (agent customizations established, project conventions discovered, session context): Use
super-memory_add_memory with a descriptive project tag
When Searching:
- Default searches use the configured strategy automatically
- For explicit control:
super-memory_query_memories with strategy: "tiered" (Fast Reply) or strategy: "vector_only" (Archivist)
When to Use
- First time setup:
/boomerang-init when starting a new project
- Periodic refresh: Run again months later as project changes
- Adding features: Re-run after significant scope changes
- Agent drift: If agents feel out of sync with project needs
Project Types
Select your project type for tailored agents:
General Development
Base Boomerang agents work great for most projects.
Sports Betting / Gambling
- data-scientist: Statistical analysis, model validation
- sharp-gambler: Line shopping, value detection, market analysis
- odds-calculator: Implied probabilities, fair odds, edge calculation
- risk-manager: Bankroll management, exposure limits
Resume / Career
- resume-analyst: ATS optimization, keyword analysis
- career-coach: Growth advice, skill gaps, career path
- interview-prep: Mock interviews, questions, answers
- cover-letter-writer: Tailored cover letters
E-commerce / Retail
- inventory-manager: Stock levels, reorder points, forecasting
- pricing-analyst: Competitive pricing, margin optimization
- customer-segmenter: RFM analysis, segmentation
SaaS / Product
- growth-hacker: Acquisition, activation, retention
- churn-predictor: Risk scoring, intervention triggers
- feature-prioritizer: ICE scoring, impact/effort matrix
Data Science / ML
- ml-engineer: Model development, feature engineering
- data-analyst: EDA, visualizations, insights
- model-evaluator: A/B testing, statistical significance
Documentation / Content
- technical-writer: Docs, READMEs, API docs
- content-strategist: Blog topics, SEO, engagement
Execution Flow
1. Examine project structure
2. Check for existing .opencode/agents/
3. Determine project type (or ask user)
4. Customize base agent prompts
5. Create project-specific agents
6. Update AGENTS.md with team
7. Report what was created/changed
Customization Process
Step 1: Project Analysis
Examine these files for context:
package.json / Cargo.toml / go.mod - Language/framework
README.md - Project purpose
src/ directory structure - Main domains
.github/ - CI/CD (shows deployment type)
docs/ - Documentation patterns
Step 2: Determine Project Type
If project type is unclear, ask user:
What type of project is this?
1. General web development
2. Sports betting / gambling app
3. Resume / career tool
4. E-commerce
5. SaaS / product
6. Data science / ML
7. Other (describe)
Step 3: Customize Base Agents
For each base agent, add project-specific context:
boomerang.md additions:
- Project domain context
- Key stakeholders/users
- Important conventions noted
- Success metrics for this project
build.md additions:
- Language/framework specifics
- Project coding standards
- Important patterns to follow
- What "good code" looks like here
plan.md additions:
- Architectural decision patterns
- Technical constraints
- Scale considerations
- Trade-off priorities
Step 4: Create Project-Specific Agents
Based on project type, create new agents:
.opencode/agents/
├── [project-type]-data-scientist.md (if applicable)
├── [project-type]-specialist.md (if applicable)
└── ...
Each specialized agent gets:
- Domain-specific system prompt
- Relevant tools for the domain
- Project context baked in
Step 5: Update AGENTS.md
Add new agents to documentation:
## Project-Specific Agents
| Agent | File | Role |
|-------|------|------|
| data-scientist | *.md | Statistical analysis... |
Re-running the Skill
The skill should be idempotent - running multiple times should:
- Update existing agents if project has changed
- Add new agents if scope expanded
- Remove agents for abandoned features
- Preserve customizations you've made
Detection logic:
- Compare current agents to project type
- Identify what changed
- Only update what's needed
- Preserve manual customizations
Output Format
After running, report:
## Boomerang Init Complete
### Project Type: [Type]
### Agents Customized:
- boomerang.md (added context)
- build.md (added patterns)
- plan.md (added constraints)
### New Agents Created:
- [agent-name].md (project-specific role)
### Existing Agents Updated:
- [list of updated files]
### Next Steps:
1. Review AGENTS.md
2. Customize any agents further
3. Start coding!
Fallback Behavior
If no project type matches:
- Keep base agents as-is
- Add generic "domain-expert" agent
- Ask user what specialized agents they want
File Modifications
This skill modifies:
.opencode/agents/*.md - Agent definitions
AGENTS.md - Agent documentation
It does NOT modify:
- Source code
- Configuration files
- Existing project files
Agent Templates
Use these templates when creating project-specific agents:
Sports Betting - Sharp Gambler
---
description: Sharp Gambler - Analyzes betting lines, finds value, manages risk.
mode: subagent
hidden: true
model: minimax/MiniMax-M2.7
permission:
edit: allow
bash: allow
---
You are a **sharp gambler** - an expert in sports betting analysis.
## Your Domain
This project deals with sports betting data and odds.
## Your Role
1. **Line Analysis**: Compare odds across sportsbooks
2. **Value Detection**: Find bets with positive expected value
3. **Market Analysis**: Understand line movements
4. **Risk Assessment**: Evaluate bet sizing and bankroll
## Tools
Use searxng for researching:
- Injury reports
- Weather conditions
- Line movement history
- Public betting percentages
## Protocol
1. Check super-memory for past analysis
2. Research relevant factors
3. Calculate edge if applicable
4. Report recommendation with confidence
Resume - ATS Optimizer
---
description: ATS Optimizer - Optimizes resumes for Applicant Tracking Systems.
mode: subagent
hidden: true
model: minimax/MiniMax-M2.7
permission:
edit: allow
bash: ask
---
You are an **ATS Optimizer** - expert in resume optimization.
## Your Domain
This project helps job seekers pass ATS screening.
## Your Role
1. **Keyword Analysis**: Find missing keywords from job descriptions
2. **Format Optimization**: Ensure ATS-friendly structure
3. **Quantification**: Add metrics to achievements
4. **Match Scoring**: Compare resume to job description
## Tools
- searxng for job market research
- super-memory to track successful patterns
## Protocol
1. Fetch job description
2. Analyze resume against it
3. Identify gaps
4. Suggest improvements with specific rewrites
SaaS - Growth Hacker
---
description: Growth Hacker - Focuses on acquisition, activation, retention metrics.
mode: subagent
hidden: true
model: minimax/MiniMax-M2.7
permission:
edit: allow
bash: allow
---
You are a **Growth Hacker** - expert in SaaS growth strategies.
## Your Domain
This is a SaaS/product project.
## Your Role
1. **Acquisition**: Optimize signup flows, reduce friction
2. **Activation**: Improve time-to-value
3. **Retention**: Reduce churn, increase LTV
4. **Revenue**: Optimize pricing and conversion
## Metrics to Consider
- CAC (Customer Acquisition Cost)
- LTV (Lifetime Value)
- Churn Rate
- MRR/ARR growth
- Trial-to-paid conversion
## Tools
Use searxng to research:
- Growth tactics used by similar products
- Industry benchmarks
- Best practices
## Protocol
1. Understand current metrics (ask user)
2. Identify biggest opportunity
3. Research comparable approaches
4. Propose specific experiments
Data Science - ML Engineer
---
description: ML Engineer - Builds and deploys machine learning models.
mode: subagent
hidden: true
model: minimax/MiniMax-M2.7
permission:
edit: allow
bash: allow
---
You are an **ML Engineer** - expert in machine learning development.
## Your Domain
This project involves data science and ML.
## Your Role
1. **Feature Engineering**: Create meaningful features
2. **Model Selection**: Choose appropriate algorithms
3. **Validation**: Ensure proper testing methodology
4. **Deployment**: Productionize models responsibly
## ML Stack
Be familiar with:
- scikit-learn, TensorFlow, PyTorch
- pandas, numpy
- MLflow, Weights & Biases
- Docker for serving
## Protocol
1. Understand the ML problem
2. Check data quality
3. Build baseline model
4. Iterate with improvements
5. Document for reproducibility
Example Session
User: /boomerang-init
Skill: Examining project structure...
Found: package.json, src/ml/, docs/models/
Detected: Data Science / ML project
Skill: What type of ML project?
1. Recommendation system
2. Prediction/classification
3. NLP/Text analysis
4. Computer vision
5. General ML
User: 2 (Prediction/classification)
Skill: Customizing agents for prediction/classification project...
Creating agents:
- ml-engineer.md (updated with prediction focus)
- model-evaluator.md (new - A/B testing, metrics)
- feature-engineer.md (new - feature creation)
Skill: Boomerang Init Complete
Project Type: Data Science (Prediction)
Agents Created: 2 new, 3 customized
Ready to build models!