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 β how one agent formally delegates a task to another β 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 β 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 β never replace β existing standards:
βββββββββββββββββββββββββββββββββββββββββββββββββββ
β Your Application β
βββββββββββββββββββββββββββββββββββββββββββββββββββ€
β Brief Protocol (.brief.md / .response.md) β β What to do
βββββββββββββββββββββββββββββββββββββββββββββββββββ€
β A2A Protocol (Transport & Discovery) β β How to find & talk
βββββββββββββββββββββββββββββββββββββββββββββββββββ€
β MCP Protocol (Tool Invocation) β β How to use tools
βββββββββββββββββββββββββββββββββββββββββββββββββββ€
β Agent Skills (SKILL.md) β β What agents know
βββββββββββββββββββββββββββββββββββββββββββββββββββ
| 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 (AβBβCβD) | Built-in depth control | Manual wiring | Not possible |
| Fan-out (AβBβ₯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 β 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** β merge after addressing minor issues.`,
});
// response.content is a valid .response.md stringconst parentBrief = Brief.parse(receivedMarkdown);
// Create a sub-brief β 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.idimport { 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: "2026-02-06T03:55:00Z"
---
# 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: "2026-02-06T04:30:00Z"
---
# 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 β B delegation
pnpm example:simple
# Cascading A β B β C chain
pnpm example:cascade
# Fan-out A β (B β₯ C) with fan-in
pnpm example:fanoutcd packages/brief-sdk
pnpm test β __tests__/parser.test.ts (9 tests)
β __tests__/brief.test.ts (10 tests)
β __tests__/response.test.ts (7 tests)
β __tests__/trace.test.ts (11 tests)
Test Files 4 passed (4)
Tests 37 passed (37)
brief/
βββ packages/brief-sdk/ # Core TypeScript SDK
β βββ src/
β β βββ brief.ts # Brief class (create, parse, sub-delegate)
β β βββ response.ts # BriefResponse class
β β βββ trace.ts # Trace audit log
β β βββ parser.ts # Markdown + YAML frontmatter parser
β β βββ validator.ts # Schema validation
β β βββ types.ts # TypeScript type definitions
β β βββ index.ts # Public API exports
β βββ __tests__/ # 37 unit tests
βββ examples/
β βββ 01-simple-delegation/ # A β B
β βββ 02-cascade-chain/ # A β B β C
β βββ 03-fan-out/ # A β (B β₯ C)
βββ docs/ # Design documents & competitive analysis
β βββ design.md # Protocol design (EN)
β βββ design-zh.md # Protocol design (ZH)
β βββ competitive-analysis.md # Market analysis (EN)
β βββ competitive-analysis-zh.md # Market analysis (ZH)
βββ README.md
We welcome contributions! Please see CONTRIBUTING.md for guidelines.
Brief β Because agents deserve proper briefings, not vague handoffs.