feat(core): add multi-agent patterns

[sharananurag998]

Summary

• add multi-agent helpers for sequential, parallel, coordinator, and redundant pipelines
• re-export new helpers from core index for external use
• prompt evaluator in parallel redundant pipeline to choose the best response

Testing

• pnpm test (fails: PostgreSQL artifact storage connection refused)

---

https://chatgpt.com/codex/tasks/task_e_68bea714b4688327820c3eae3b2143ad

[Copilot]

Pull Request Overview

This PR adds multi-agent orchestration patterns to the JAF core library, introducing helper functions for sequential, parallel, coordinator, and redundant pipelines. The changes enable developers to compose multiple agents in sophisticated workflows beyond simple single-agent interactions.

• Implements four multi-agent patterns: sequential pipelines, parallel pipelines, coordinator routing, and parallel redundant evaluation
• Exports new multi-agent functions from core index for external use
• Creates comprehensive demo example with mock model provider to showcase all patterns

Reviewed Changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/index.ts	Exports multi-agent module from core
src/core/multi-agent.ts	Implements four multi-agent orchestration patterns
examples/multi-agent-demo/tsconfig.json	TypeScript configuration for demo project
examples/multi-agent-demo/package.json	Package configuration with workspace dependency
examples/multi-agent-demo/index.ts	Demo showcasing all multi-agent patterns with mock agents
examples/multi-agent-demo/README.md	Documentation explaining demo usage and patterns

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[Copilot]

Using non-null assertion operator on lastResult is unsafe. If agentNames is empty, lastResult will be undefined and this will throw a runtime error. Consider adding a guard clause or returning an appropriate error result for empty arrays.

[Copilot]

[nitpick] The output combination logic uses hardcoded newline separation. Consider making the separator configurable or using a more structured approach to combine outputs, as different use cases may require different formatting.

[Copilot]

[nitpick] The Judge behavior relies on hardcoded string matching which is brittle. This makes the demo fragile to changes in the evaluation input format. Consider using a more robust parsing approach or making the expected format explicit.

Suggested change

	Judge: input => (input.includes('Response from Answer1') ? '42' : '41')
	Judge: input => {
	// Parse responses from Answer1 and Answer2
	// Expected format: "Response from Answer1: <answer1>\nResponse from Answer2: <answer2>"
	const answer1Match = input.match(/Response from Answer1: ([^\n]+)/);
	const answer2Match = input.match(/Response from Answer2: ([^\n]+)/);
	const answer1 = answer1Match ? answer1Match[1].trim() : null;
	const answer2 = answer2Match ? answer2Match[1].trim() : null;
	// Prefer answer1 if present, otherwise answer2
	if (answer1) return answer1;
	if (answer2) return answer2;
	// Fallback if neither found
	return 'No valid answer found';
	}