A narrative handbook on the agent harness — the tooling, context management, permissions, and feedback loops wrapped around a raw language model — and how two coding agents build it differently.
Read in order
The harness is everything wrapped around the raw model: the tool interface, context management, permissions and sandboxing, and verification loops. Why the harness — not just the model — decides agent quality.
Built-in tools, Agent Skills and progressive disclosure, subagents and the Task tool, hooks, plugins and marketplaces, permission modes, CLAUDE.md memory, and MCP.
AGENTS.md project instructions, approval modes and OS-level sandboxing, config.toml, MCP support, and how the open-source Rust agent manages context and runs commands.
A balanced comparison across context model, extensibility, permissions, multi-agent, and ecosystem — honest trade-offs and practical guidance on picking one.