Compatibility Matrix

Capability coverage across harnesses and adapters

✔ Native support
⚠ Partial support
✘ Not supported
🎯 Planned

Adapter Availability

Adapter Package Status Notes
Anthropic Agent SDK @openharness/adapter-anthropic-agent ✔ Tool use, streaming, conversations
Claude Code CLI - 🎯 No public API
Goose - 🎯 Planned
LangChain Deep Agent - 🎯 Planned
Letta - 🎯 Planned

Domain-Level Support

Domain Anthropic Agent Claude Code Goose Deep Agent Letta
Agents ✘ ⚠ ⚠ ⚠ ✔
Skills ✘ ✔ ✔ ⚠ ⚠
Tools ✔ ✔ ✔ ✔ ✔
MCP ✘ ✔ ✔ ⚠ ⚠
Execution ✔ ✔ ✔ ✔ ✔
Sessions ✔ ⚠ ✔ ⚠ ✔
Memory ✘ ⚠ ⚠ ⚠ ✔
Subagents ✘ ⚠ ✘ ✔ ✘
Files ✘ ✔ ✔ ✔ ⚠
Hooks ✘ ✔ ⚠ ⚠ ⚠
Planning ✘ ⚠ ⚠ ✔ ⚠
Models ⚠ ✘ ✔ ✔ ✔

Detailed Operations

Execution

Operation Anthropic Agent Claude Code Goose Deep Agent Letta
Sync ✔ ✔ ✔ ✔ ✔
Stream ✔ ✔ ✔ ✔ ✔
Cancel ✔ ✔ ⚠ ⚠ ⚠
Tool Calls ✔ ✔ ✔ ✔ ✔
Extended Thinking ✔ ✔ ✘ ✘ ✘
Artifacts ✘ ✔ ✔ ✔ ⚠

Tools

Operation Anthropic Agent Claude Code Goose Deep Agent Letta
Register ✔ ✔ ✔ ✔ ✔
Unregister ✔ ⚠ ⚠ ⚠ ⚠
List ✔ ✔ ✔ ✔ ✔
Invoke ✔ ✔ ✔ ✔ ✔

Sessions

Operation Anthropic Agent Claude Code Goose Deep Agent Letta
Create ✔ ⚠ ✔ ⚠ ✔
Resume ✘ ⚠ ✔ ⚠ ✔
History ✔ ⚠ ✔ ⚠ ✔
Delete ✔ ⚠ ✔ ⚠ ✔
Fork ✘ ✘ ✘ ✘ ⚠

MCP

Operation Anthropic Agent Claude Code Goose Deep Agent Letta
Connect ✘ ✔ ✔ ⚠ ⚠
Disconnect ✘ ✔ ✔ ⚠ ⚠
List Tools ✘ ✔ ✔ ⚠ ⚠
Resources ✘ ✔ ✔ ✘ ✘
Prompts ✘ ✔ ✔ ✘ ✘

Harness Profiles

Anthropic Agent SDK

@openharness/adapter-anthropic-agent
Strengths
  • Direct Anthropic Messages API access
  • Native streaming with async generators
  • Tool use with automatic agentic loop
  • Extended thinking support
  • Cancellation via AbortSignal
Limitations
  • In-memory only (no persistence)
  • Single provider (Anthropic)
  • No MCP support built-in
  • No file operations

Claude Code (CLI)

Planned
Strengths
  • Native skills with API registration
  • Full MCP support
  • Comprehensive hooks system
  • Strong file operations
Limitations
  • No public programmatic API
  • Single provider (Anthropic)
  • Agent lifecycle management

Goose

Planned
Strengths
  • MCP-first architecture
  • Multi-model support (25+ providers)
  • Strong session management
  • Native skills support
Limitations
  • No subagent spawning
  • Persistent memory limited
  • Skill versioning API

LangChain Deep Agent

Planned
Strengths
  • Built-in planning/todos
  • Native subagent delegation
  • Flexible filesystem backends
  • Multi-model support
Limitations
  • Skills API (uses tools)
  • Full MCP integration
  • Hooks system

Letta

Planned
Strengths
  • Memory-first architecture
  • Full agent lifecycle API
  • Cross-session persistence
  • Multi-model support
Limitations
  • Native skills support
  • Subagent delegation
  • File operations
  • MCP integration

Coverage Statistics

24%
Anthropic Agent SDK
59%
Claude Code
51%
Goose
47%
Deep Agent
51%
Letta

Coverage = (Supported + 0.5 × Partial) / Total Operations