Deep Code Reader

Systematically read and understand a codebase, producing a set of verified cognitive skills that capture deep knowledge — module capabilities, design logic, data structures, state flow, and modification guides.

The core mechanism: a closed-book exam verification loop ensures generated skills are genuinely comprehensive, not shallow summaries.

1. The Team Roles

To make this process robust and conceptually clear, the system employs three distinct agents modeled after a software engineering team:

• Agent A (Tech Writer): The deep reader. Reads the source code and writes the comprehensive skill document.
• Agent B (QA Engineer): The examiner. Reads the source code, extracts verifiable facts, and generates test questions.
• Agent C (Junior Dev): The candidate. Acts as a new team member who can ONLY read the document written by Agent A to answer Agent B's questions.

2. Usage

Here is the CLI command to trigger the deep-code-read workflow: