Claude Teams Language Protocol

AgentSpeak v2: Token-Efficient Protocol for Claude Code Agent Teams

**Cut inter-agent communication tokens by 60-70%** with a structured, compressed language protocol for [Claude Code Agent Teams](https://code.claude.com/docs/en/agent-teams).

The Problem

When Claude Code agents communicate in a team, each message consumes tokens. With 4-5 agents messaging each other, token costs scale fast. Default English messages like:

"I've finished fixing the VideoFeed component. I removed 12 console.log statements and changed the translateY animation to use CSS scroll-snap instead. TypeScript compiles cleanly."

...cost **~40 tokens**. The same message in AgentSpeak:

`deltaT1 VF ~translateY->scroll-snap -12log. tsc clean.`

...costs **~15 tokens**. Multiply across hundreds of inter-agent messages per session.

What's New in v2

AgentSpeak v2 integrates with Claude Code's native [Agent Teams](https://code.claude.com/docs/en/agent-teams) system, bridging the protocol with shared task lists, hooks, and dependency tracking.

Feature	v1	v2
Status codes + actions	Yes	Yes
File shortcodes	Yes	Yes
Task ID references (T#)	--	Maps to shared task list
Dependency notation (>> / <<)	--	Unblock/block signals
File locking (+lock / -lock)	--	Runtime ownership claims
Message priority (!! / ..)	--	Urgent / normal / FYI
Progress indicator (beta75)	--	Quantitative % done
Lead routing shorthand (@agent [])	--	Compact assignment
QA batch format (epsilonBATCH)	--	Structured severity groups
Error recovery (epsilon!retry)	--	Retry/fail/reassign
Hook integration (HOOK:verify)	--	TeammateIdle + TaskCompleted
Protocol versioning (PROTO:v2.0)	--	Version declaration at spawn
Protocol feedback (theta)	--	Agent-driven improvement PRs

Quick Start

**New to agent teams?** Read the [**Qui