127 lines
4.4 KiB
Markdown
127 lines
4.4 KiB
Markdown
---
|
|
description: "Parallelizes tasks with specialized agents"
|
|
prerequisites: "—"
|
|
argument-hint: "<task_description> [--workers=N] [--strategy=auto|error|test|lint|api|database|type|import]"
|
|
allowed-tools: ["Task", "TodoWrite", "Glob", "Grep", "Read", "LS", "Bash", "SlashCommand"]
|
|
---
|
|
|
|
Parallelize the following task using specialized agents: $ARGUMENTS
|
|
|
|
## Task Analysis
|
|
|
|
Parse the arguments to understand what specialized agents are needed:
|
|
|
|
- Extract any `--workers=N` or `--strategy=TYPE` options
|
|
- Analyze the task content to detect which domain expertise is required
|
|
- Identify the core work and how it can be distributed
|
|
|
|
## Specialized Agent Detection
|
|
|
|
Determine which specialized agent types would be most effective:
|
|
|
|
**Error-focused agents:**
|
|
|
|
- `type-error-fixer` - For mypy errors, TypeVar, Protocol, type annotations
|
|
- `import-error-fixer` - For ModuleNotFoundError, import issues, Python path problems
|
|
- `linting-fixer` - For ruff, format issues, E501, F401 violations
|
|
- `api-test-fixer` - For FastAPI, endpoint tests, HTTP client issues
|
|
- `database-test-fixer` - For database connections, fixtures, SQL, Supabase issues
|
|
- `unit-test-fixer` - For pytest failures, assertions, mocks, test logic
|
|
|
|
**Workflow agents:**
|
|
|
|
- `commit-orchestrator` - For git commits, staging, pre-commit hooks, quality gates
|
|
- `ci-workflow-orchestrator` - For CI/CD failures, GitHub Actions, pipeline issues
|
|
|
|
**Investigation agents:**
|
|
|
|
- `digdeep` - For root cause analysis, mysterious failures, complex debugging
|
|
- `security-scanner` - For vulnerabilities, OWASP compliance, secrets detection
|
|
- `performance-test-fixer` - For load tests, response times, benchmarks
|
|
- `e2e-test-fixer` - For end-to-end workflows, integration tests
|
|
|
|
**Fallback:**
|
|
|
|
- `parallel-executor` - For general independent parallel work
|
|
- `general-purpose` - For complex multi-domain coordination
|
|
|
|
## Work Package Creation
|
|
|
|
Use available tools to understand the codebase and create specialized work packages:
|
|
|
|
- Use LS to examine project structure
|
|
- Use Grep to identify error patterns or relevant files
|
|
- Use Read to examine error outputs or test results
|
|
|
|
Then divide the task by domain expertise:
|
|
|
|
**Single-domain tasks** (e.g., "fix all linting errors"):
|
|
|
|
- Create 1-2 work packages for the same specialized agent type
|
|
- Group by file or error type
|
|
|
|
**Multi-domain tasks** (e.g., "fix test failures"):
|
|
|
|
- Analyze test output to categorize failures by type
|
|
- Create one work package per error category
|
|
- Assign appropriate specialized agent for each category
|
|
|
|
**Mixed-strategy tasks**:
|
|
|
|
- Categorize issues by required domain expertise
|
|
- Create specialized work packages for each agent type
|
|
- Ensure no overlap in file modifications
|
|
|
|
## Agent Execution
|
|
|
|
Launch multiple specialized Task agents in parallel (all in a single message) using the appropriate `subagent_type`.
|
|
|
|
**Best practices:**
|
|
|
|
- Send all Task tool calls in one batch for true parallelization
|
|
- Match agent type to problem domain for higher success rates
|
|
- Give each agent clear domain-specific scope
|
|
- Ensure agents don't modify the same files
|
|
|
|
**Agent specialization advantages:**
|
|
|
|
- Domain-specific tools and knowledge
|
|
- Optimized approaches for specific problem types
|
|
- Better error pattern recognition
|
|
- Higher fix success rates
|
|
|
|
Each specialized agent prompt should include:
|
|
|
|
- The agent's domain expertise and role
|
|
- Specific scope (files/directories/error types to address)
|
|
- The specialized work to complete
|
|
- Constraints to avoid conflicts with other agents
|
|
- Expected output format including cross-domain issues
|
|
|
|
## Result Synthesis
|
|
|
|
After specialized agents complete:
|
|
|
|
- Validate each agent's domain-specific results
|
|
- Identify any cross-domain conflicts or dependencies
|
|
- Merge findings into a coherent summary
|
|
- Report which agent types were most effective
|
|
- Recommend follow-up work if issues require different specializations
|
|
|
|
## Quick Reference: Agent Type Mapping
|
|
|
|
- **Linting** → `linting-fixer`
|
|
- **Type errors** → `type-error-fixer`
|
|
- **Import errors** → `import-error-fixer`
|
|
- **API tests** → `api-test-fixer`
|
|
- **Database tests** → `database-test-fixer`
|
|
- **Unit tests** → `unit-test-fixer`
|
|
- **Git commits** → `commit-orchestrator`
|
|
- **CI/CD** → `ci-workflow-orchestrator`
|
|
- **Deep investigation** → `digdeep`
|
|
- **Security** → `security-scanner`
|
|
- **Performance** → `performance-test-fixer`
|
|
- **E2E tests** → `e2e-test-fixer`
|
|
- **Independent tasks** → `parallel-executor`
|
|
- **Complex coordination** → `general-purpose`
|