Better Harness: setup and workflow audit in Claude Code

This Portuguese-language tutorial demonstrates Better Harness in a Claude Code project. The speaker presents the plugin as a way to audit an AI agent's working context, rules and validation practices, arguing that these can limit results independently of the model. That is the speaker's explanation, rather than a demonstrated guarantee of better performance across agents.

The installation walkthrough adds a marketplace, installs the plugin for the project and reloads plugins. The speaker then builds Comanda Rápida, a sample ordering system for small restaurants, before requesting a quick audit. Better Harness launches three parallel subagents to examine project context, configuration assets and sessions, then produces a report with recommendations.

The report flags missing version control, a lack of automated checks for bill calculations and insufficient project documentation. The speaker copies suggested repair prompts back into Claude Code to update documentation and address calculation verification. The example shows how a coding assistant can use an audit report to guide follow-up work.

The demonstration also exposes a limitation: the report finds zero sessions despite existing conversations. The speaker attributes this to underscores in the project folder name breaking session discovery and says it affected the score. The results therefore include an incomplete session analysis.