Multi-Harness Coding Prototype: OpenCode, Pi, and Large Repositories

One-sentence takeaway

Aizzy's public suggestion is to connect Ox Alpha to OpenCode, Pi, or other coding harnesses and test long tasks on a redacted large repository; the reusable part is fixing the harness, task, and logs rather than repeating the marketing claim.

Use cases

• Suitable tasks: Comparing coding-agent harnesses, repository understanding, and prototype application development.

• Unsuitable tasks: Equating platform privacy or free-trial claims with every provider's terms, or uploading a sensitive repository without review.

• Applicable model version: Ox Alpha; no version snapshot is provided.

• Applicable clients, agents, or APIs: OpenCode, Pi, and other coding harnesses named in the source.

• Recommended reasoning levels and parameters: Not disclosed; fix context, tool permissions, concurrency, and timeouts separately for each harness.

Ready-to-use content

The source does not publish a verbatim prompt. The following workflow is derived from its public recommendation:

1. Choose a redacted, reversible repository. Fix the commit and test command.
2. Connect Ox Alpha through OpenCode, Pi, and one other harness; change only the harness between runs.
3. First ask the agent to inspect the repository and state a plan, then execute one verifiable coding task.
4. Save the complete tool trace, model output, errors, duration, tokens, and final diff.
5. Run tests and review side effects manually. Report task completion, rework, and repeated output separately.
6. If the model switches languages, becomes lazy, or retries indefinitely, preserve the raw trace instead of replacing it with a summary.

Original information and boundaries

Aizzy's other public post says that after testing a large codebase the model was “good so far” but “a bit lazy,” and sometimes replied in Russian. These are personal observations. The account also promotes its platform, so its privacy and free-access claims should be checked against the current terms.