skills/parallel-agent-contracts/SKILL.md
Parallel Agent Type Contracts
npx skillsauth add rubicanjr/FinCognis parallel-agent-contractsInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
When launching parallel agents for code implementation, prevent type duplication.
## Before Marking Complete
Run verification:
\`\`\`bash
npx tsc --noEmit 2>&1 | head -20
\`\`\`
If ANY type errors exist, fix them before completing.
## Before Creating Any Type/Interface
First check if it exists:
\`\`\`bash
grep -r "interface YourTypeName\|type YourTypeName" src/
\`\`\`
If found, import it. NEVER duplicate existing types.
Include relevant entries from this map in agent prompts:
| Type | Owner File | Import From |
|------|-----------|-------------|
| NormalizedTool | src/sdk/agent.ts | './agent' |
| ToolCall | src/sdk/agent.ts | './agent' |
| ToolResult | src/sdk/agent.ts | './agent' |
| ToolDefinition | src/sdk/agent.ts | './agent' |
| Message | src/sdk/types.ts | './types' |
| ContentBlock | src/sdk/types.ts | './types' |
| TokenUsage | src/sdk/types.ts | './types' |
| ProviderAdapter | src/sdk/providers/index.ts | './providers' |
| RiggClient | src/sdk/client.ts | './client' |
When spawning implementation agents:
# Task: [Description]
## Type Ownership (DO NOT recreate)
- [List relevant types from canonical map]
## Before Creating New Types
Run: `grep -r "interface TypeName" src/` - if exists, import it.
## Before Marking Complete
Run: `npx tsc --noEmit 2>&1 | head -20`
Fix all type errors before completing.
## Your Implementation
[Actual task description]
development
Goal-based workflow orchestration - routes tasks to specialist agents based on user goals
tools
Wiring Verification
development
Connection management, room patterns, reconnection strategies, message buffering, and binary protocol design.
development
Screenshot comparison QA for frontend development. Takes a screenshot of the current implementation, scores it across multiple visual dimensions, and returns a structured PASS/REVISE/FAIL verdict with concrete fixes. Use when implementing UI from a design reference or verifying visual correctness.