Transform a rough idea into a comprehensive implementation plan. Two-stage process: improve the prompt via codebase reconnaissance, then generate a detailed phased plan with atomic checkboxes, file paths, and signatures.

Comprehensive codebase health assessment across 5 quantitative dimensions: concept scattering, module depth vs. width, test health, coupling risk, and AI-agent readability. Produces a letter-grade scorecard.

Commit and push all uncommitted changes. Discovers modified and untracked files, generates diffs, drafts a descriptive commit message with bullet points, and presents it for approval before committing and pushing.

Commit and push only files modified by the AI agent across the current session. Scans the conversation transcript to identify edited files and cross-references with git status to detect prior session changes.

Rigorous structured interview to stress-test an idea before building. Conducts themed rounds (intent, users, data, edges, integration) grounded in codebase reality. Produces a Decision Record.