Open Harness

The description states that Open Harness is a local-first, open-source control plane for governed GPT-5.6 agent workflows. It is built as a TypeScript/Bun monorepo, with three main packages:

• harness-core: contains routing, safe configuration lifecycle, runtime discovery, background-session controls, permissions, worktree coordination, checkpoints, and audit logic.
• opencode-plugin: connects lifecycle events, routing, telemetry, notifications, and compaction to OpenCode.
• cli: provides installation, diagnostics, model status, lifecycle operations, demonstrations, and explicitly gated live checks.

It is described as a runtime discovery system that routes work among logical model classes (Sol, Terra, Luna) and supports transactional install, upgrade, migration, and ownership-aware uninstall, with typed child-session controls for spawn, wait, inspect, steer, cancel, collect, fork, and revert.

It also includes:

• Git-worktree-aware writer contracts
• Bounded resumable checkpoints that store identifiers and lifecycle metadata rather than prompts or credentials
• Local, content-free audit and usage records
• Quota-gated live verification paths
• Cross-platform release targets (Darwin arm64 verified locally)

The project does not read or import Codex credentials. OpenCode remains the execution engine and credential owner.

Inference: The system is designed to manage agent workflows in a secure, auditable, and resumable way, with emphasis on local-first execution and governance.

The description states that Open Harness explores a more transparent answer: an open-source, local-first orchestration and governance layer for OpenCode. It positions itself as a solution to the fragmented control plane in agent coding tools, where model selection, permissions, worktree isolation, resumability, auditability, and safe installation are often left to prompt conventions or one-off scripts.

It claims to make the GPT-5.6 model family usable as a coherent engineering team, by routing tasks to models based on their strengths (Sol for architecture, Terra for bounded implementation, Luna for verification).

The project also states that it was built using Codex and GPT-5.6 to orchestrate and verify the development process itself — splitting tasks into read-only architecture, platform, live-session, and observability investigations before allowing a single isolated writer to modify source code.

Inference: The positioning is that of a secure, local-first control plane for managing agent workflows, with a focus on governance, auditability, and safe execution. It is not positioned as a commercial product but as an experimental tool for developers working with OpenCode and GPT-5.6.

The description does not state the target customer or ideal customer profile (ICP). It implies that the intended users are developers working with OpenCode and GPT-5.6, particularly those who want to orchestrate agent workflows in a secure, auditable way.

It is described as a tool for engineering teams who need to manage multiple agents in parallel, with concerns around permissions, resumability, auditability, and safe installation.

Inference: The ICP appears to be technical users or developers working with AI agents, especially those using OpenCode and GPT-5.6 models, but no explicit customer segment is defined.

There is no evidence of a business model or pricing structure in the description. The project is described as open-source (MIT licensed), local-first, and built for developers to use in their own environments.

The author states that Open Harness does not read or import Codex credentials, and OpenCode remains the execution engine and credential owner — suggesting no direct monetization of the tool itself.

Inference: No business model or pricing is evident. The project is presented as a developer tool, not a commercial offering.

The project is built using:

• TypeScript
• Bun
• Git
• OpenCode
• GPT-5.6 models (Sol, Terra, Luna)

It includes:

• A monorepo structure with three main packages
• Transactional install and upgrade mechanisms
• Runtime model discovery instead of hard-coded assumptions
• Typed child-session controls
• Git-worktree-aware writer contracts
• Bounded resumable checkpoints
• Local, content-free audit records
• Quota-gated live verification paths
• Cross-platform release targets (Darwin arm64 verified)

It also includes:

• 53 automated tests across various domains
• TypeScript and source-format checks
• Security policy check passes
• Compiled release installs and smoke tests in temporary workspace
• Live model tests are plan-only by default, requiring explicit quota consent

Inference: The technical architecture is developer-focused, with strong emphasis on security, auditability, and local execution. It is a prototype or early-stage tool, not a production-ready product.

The description states that this is a hackathon submission (OpenAI 2026 hackathon). There is no evidence of:

• Revenue
• Customers
• Adoption
• Product-market fit
• Usage beyond the author’s own environment

It is described as a tested local control-plane foundation, but there is no indication that it has been used in production or by others.

Inference: No traction or maturity signals are evident. The project is at an early prototype stage, likely not yet in use by anyone other than the author.

The description does not mention any competitors. It is positioned as a solution to fragmented control planes in agent coding tools, but no specific market players or products are named.

It is described as a local-first, open-source tool for managing GPT-5.6 workflows, which suggests it may compete with or complement existing AI agent orchestration platforms — though no such platforms are identified.

Inference: No competitive context is provided. The project appears to be independent of known market players, and its positioning is not compared to existing tools in the space.

• No evidence of traction or adoption: This is a hackathon submission with no indication of real-world usage.
• Self-reported only: All claims are unverified, and there is no third-party validation.
• No commercialization path: The project is open-source and local-first — no clear monetization strategy.
• Limited scope: It appears to be a developer tool, not a product for end-users or enterprises.
• Unproven in production: No evidence of use beyond the author’s own environment.

Inference: The project is at a very early stage, with no commercial viability or market traction evident. It may be a proof-of-concept or experimental tool, not a product ready for adoption.

1. What specific use cases are you targeting with this tool? Is there any real-world testing beyond the author’s own environment?
2. How does Open Harness integrate with existing workflows in development teams using OpenCode?
3. Are there plans to expand beyond the current local-first, open-source model into enterprise or commercial offerings?
4. What is the long-term roadmap for the project, and how will it evolve from a hackathon prototype?
5. Have you considered any potential security or compliance risks with the checkpointing and audit mechanisms?