What makes these agents specific to Claude Code?

[proofer]

The question is in the title.

[wshobson]

These agents are highly specific to Claude Code.

Key Claude Code-Specific Features:

1. Native Task Tool: Only Claude Code has built-in agent delegation via Use Task tool with subagent_type="backend-architect". Other tools lack this native mechanism.
2. Unique Agent Format: Claude Code uses YAML frontmatter in markdown files. Others use completely different formats (GitHub Copilot: JSON APIs, Cursor: .cursorrules, Continue: config.json with MCP).
3. Directory-Based Discovery: Agents auto-load from ~/.claude/agents/. Other tools use external hosting (Copilot) or centralized config (Continue/Cursor).
4. Isolated Context Windows: Each agent gets its own context - unique to Claude Code's architecture.
5. Claude Model Names: The haiku/sonnet/opus specifications are Claude-specific.

Why They Won't Work Elsewhere:

• No Task Tool equivalent in other systems
• Different invocation patterns (APIs vs local files)
• Incompatible formats (YAML/markdown vs JSON/APIs)
• Missing delegation infrastructure

Could They Be Adapted?

Theoretically yes, but would require:

• Complete format conversion
• Building custom delegation systems
• Model name mapping
• New invocation mechanisms

[proofer]

For someone just getting into agents, this is very informative! And this repo looks to be an excellent source for implementation and inspiration. Thanks!