Running the SLOBAC Audit

Install

Install the txrk9-agent-plugins Plugin Marketplace into your harness of choice.

Then, install the slobac plugin from that marketplace.

This will give your harness access to the /slobac-audit Agent Skill.

Scope

The audit is intended to be read-only: it reports findings; it does not modify test code. A report is created that you can feed into your chosen remediation process.

You may run the audit against a single test, a single test suite, a directory tree, or multiple trees of tests within your repository - it will ask you to choose.

You may run the audit looking for a subset of the taxonomy of test smells. If you don't specify, it will ask you to choose. You can use all to search for all smells.

/slobac-audit src/test/java/ all smells

Context window

For best results, run SLOBAC with the most-capable model and largest context window available.

• In Cursor, enable MAX mode and pick a heavyweight, frontier model.
• In Claude Code, use Opus with the 1M context window.

Larger context means fewer batches, richer cross-suite analysis, and better recall on redundancy detection. SLOBAC works at 200K context but shards more aggressively, trading recall on cross-suite smells for safety.

Pass your context window size in the invocation to skip the one-time question the orchestrator asks when it encounters a large suite without a stated budget.

/slobac-audit src/test/java/ all smells - 1M context window