Turn chaotic multi-agent handoffs into structured, auditable task briefings.
The AI agent ecosystem now has established standards for tool invocation (MCP), agent discovery and transport (A2A), and agent capabilities (Agent Skills). Yet the most critical interaction — how one agent formally delegates a task to another — remains unstandardized.
The Cognition team (creators of Devin) identified this exact gap in their influential post "Don't Build Multi-Agents", calling it the "Implicit Handoff" problem: agents pass tasks without structured context, leading to decision conflicts, context loss, and fragile systems. Their diagnosis was precise. But rather than abandoning multi-agent architectures, we believe the answer is to formalize the delegation itself.
Brief is the first open protocol dedicated to solving this problem. It introduces a standardized, human-readable format for inter-agent task delegation — filling the missing semantic layer between how agents communicate (A2A) and what agents can do (MCP/Skills).
The entire protocol is two Markdown files:
Agent A: sends brief-001.brief.md
"Here's exactly what I need, why, and what I expect back."
Agent B: sends brief-001.response.md
"Here's exactly what I did and what I'm delivering."
That's it. No new DSL, no binary format, no framework lock-in. If you can write Markdown, you can use Brief.
Brief is not another framework. It is a semantic layer designed to complement — never replace — existing standards:
┌─────────────────────────────────────────────────┐
│ Your Application │
├─────────────────────────────────────────────────┤
│ Brief Protocol (.brief.md / .response.md) │ ← What to do
├─────────────────────────────────────────────────┤
│ A2A Protocol (Transport & Discovery) │ ← How to find & talk
├─────────────────────────────────────────────────┤
│ MCP Protocol (Tool Invocation) │ ← How to use tools
├─────────────────────────────────────────────────┤
│ Agent Skills (SKILL.md) │ ← What agents know
└─────────────────────────────────────────────────┘
| Feature | Brief | Ad-hoc Multi-Agent | Single Monolithic Agent |
|---|---|---|---|
| Context isolation | Explicit contract | Shared memory (polluted) | N/A |
| Auditability | Full trace log | Black box | Single log |
| Cascading (A→B→C→D) | Built-in depth control | Manual wiring | Not possible |
| Fan-out (A→B∥C) | Native support | Complex orchestration | Not possible |
| Learning curve | Write Markdown | Learn framework API | N/A |
| Ecosystem fit | Complements MCP + A2A | Replaces everything | N/A |
npm install brief-sdk
# or
pnpm add brief-sdkimport { Brief } from 'brief-sdk'; const brief = Brief.create({ delegator: 'agent-orchestrator', delegatee: 'agent-code-reviewer', body: `# Briefing: Review PR #42 ## Objective Review the OAuth 2.0 migration code for security issues. ## Key Constraints - Focus on token handling and storage. - Check for N+1 queries. ## Expected Deliverables 1. Review summary with findings. 2. Approve or reject recommendation.`, }); // brief.content is a valid .brief.md string — send it however you want console.log(brief.content);
import { Brief } from 'brief-sdk'; const incoming = Brief.parse(receivedMarkdown); console.log(incoming.delegator); // "agent-orchestrator" console.log(incoming.isSubBrief); // false // Do your work... const results = await myAgent.review(incoming.body); // Create a structured response const response = incoming.createResponse({ status: 'success', body: `# Review Complete ## Findings - Approved with minor suggestions. - Found unused import in auth/handler.ts. ## Recommendation **Approve** — merge after addressing minor issues.`, }); // response.content is a valid .response.md string
const parentBrief = Brief.parse(receivedMarkdown); // Create a sub-brief — parentId and depth are set automatically const subBrief = parentBrief.createSubBrief({ delegator: 'agent-fullstack-dev', delegatee: 'agent-docs-writer', body: '# Write API docs for the new endpoints.', }); console.log(subBrief.isSubBrief); // true console.log(subBrief.parentId); // parentBrief.id
import { Trace } from 'brief-sdk'; const trace = Trace.create(); trace.append({ agent: 'agent-orchestrator', action: 'Delegated code review.', briefId: 'brief-001', }); trace.append({ agent: 'agent-code-reviewer', action: 'Completed review. Status: success.', briefId: 'brief-001', }); // Serialize to trace.md console.log(trace.toString());
A Brief document is a standard Markdown file with YAML frontmatter:
--- id: "brief-a4b1c8" protocolVersion: "1.2.0" delegator: "agent-orchestrator" delegatee: "agent-code-reviewer" timestamp: "2026年02月06日T03:55:00Z" --- # Briefing: Your Task Title ## 1. Objective What needs to be done. ## 2. Context Why it needs to be done and relevant background. ## 3. Constraints Rules and limitations. ## 4. Expected Deliverables What to return.
--- id: "brief-a4b1c8" status: "success" timestamp: "2026年02月06日T04:30:00Z" --- # Response: Task Complete ## 1. Summary What was done. ## 2. Delivered Artifacts Links and references to outputs.
For multi-level delegation, add these optional fields to the frontmatter:
| Field | Type | Description |
|---|---|---|
parentId |
string |
Links this brief to a parent brief |
maxDepth |
number |
Maximum allowed delegation depth |
currentDepth |
number |
Current depth in the chain |
The examples/ directory contains three runnable demos:
# Simple A → B delegation pnpm example:simple # Cascading A → B → C chain pnpm example:cascade # Fan-out A → (B ∥ C) with fan-in pnpm example:fanout
cd packages/brief-sdk pnpm test
✓ __tests__/parser.test.ts (9 tests)
✓ __tests__/brief.test.ts (10 tests)
✓ __tests__/response.test.ts (7 tests)
✓ __tests__/trace.test.ts (11 tests)
Test Files 4 passed (4)
Tests 37 passed (37)
brief/
├── packages/brief-sdk/ # Core TypeScript SDK
│ ├── src/
│ │ ├── brief.ts # Brief class (create, parse, sub-delegate)
│ │ ├── response.ts # BriefResponse class
│ │ ├── trace.ts # Trace audit log
│ │ ├── parser.ts # Markdown + YAML frontmatter parser
│ │ ├── validator.ts # Schema validation
│ │ ├── types.ts # TypeScript type definitions
│ │ └── index.ts # Public API exports
│ └── __tests__/ # 37 unit tests
├── examples/
│ ├── 01-simple-delegation/ # A → B
│ ├── 02-cascade-chain/ # A → B → C
│ └── 03-fan-out/ # A → (B ∥ C)
├── docs/ # Design documents & competitive analysis
│ ├── design.md # Protocol design (EN)
│ ├── design-zh.md # Protocol design (ZH)
│ ├── competitive-analysis.md # Market analysis (EN)
│ └── competitive-analysis-zh.md # Market analysis (ZH)
└── README.md
We welcome contributions! Please see CONTRIBUTING.md for guidelines.
Brief — Because agents deserve proper briefings, not vague handoffs.