HarnessSmith

What it does

HarnessSmith is a specialized generator that produces complete, independent Python repositories for AI agent harnesses. Unlike traditional agent frameworks that lock you into a specific runtime (like LangGraph or CrewAI), HarnessSmith renders a thin, readable Python loop that you fully own and can edit or delete any line of without dependency on the generator.

Tools

• Native Function Calling: Drives models through API tool_calls with ReAct semantics.
• Dual LLM Protocol: Built-in clients for both OpenAI-compatible endpoints and native Anthropic Messages.
• Context Management: Implements trigger-based compaction (truncate/summarize) and tool-result clipping.
• Human-in-the-Loop: Includes ask_question for structured clarification and risk-based tool confirmation gates.
• Budgeting: Persistent per-LLM cost accounting with hard cost limits.

Installation

To install the generator and start the wizard:

uvx --from "harnessmith[wizard]" harnessmith wizard --open

For the generated harness, use the provided one-click launcher or:

uv run my-agent serve --open

Supported hosts

• claude