[Agentic] - Custom Agent Harnesses and Open Model's Investor Website

About the project

An agent runtime becomes much more useful when it fits the way a person actually works. My harness work connects existing engines to project context, local tools and an operator’s workspace. OpenClaw and Hermes provide the underlying frameworks. My contribution is the surrounding integration: skills and workflow conventions, status reporting, launch adapters, session identity, and the feedback that helps a person understand what an agent is doing.

Integration work

• Configurable launch aliases for different agent runtimes, including OpenClaw chat and Hermes terminal sessions.
• Project- and role-aware launch metadata that connects a visible terminal session to its coordination identity.
• Hermes startup classification from terminal output, distinguishing pending, ready and error states.
• File-backed bootstrap context for long prompts and session handoffs.
• Local agent status files with task progress, errors, history and usage fields for an operator interface.
• Runtime logging utilities for inspecting calls, timing and test-run outcomes.

How the projects connect

This entry focuses on adapting engines and extending the working environment. Swarm MCP Lab supplies coordination and session infrastructure; Haberdashery explores the operator interface. Some harness adapters live within those projects, while earlier OpenClaw skills and workflow tools live in a separate local workspace.

Current state

The local tools and adapters are implemented across the Agentic workspaces, with source and focused-test coverage for several integration paths. Native launch, permissions and cleanup need to be verified for each runtime and environment. The broader cross-platform Agentic HQ remains a developing product direction.

Built with

Gallery