interview-framework

Codebase-first discovery identifies whether needed information is a code fact, prior memory fact, current-docs fact, or a user decision, then delegates accordingly to parallel agents before asking the user for input.

An adaptive user decision interview is a structured three-phase dialogue flow—understand, propose approaches, and confirm and store—that runs after parallel pre-discovery to gather codebase facts before asking user-only decisions across spec phases.

Structure the interview by running an UNDERSTAND decision-tree phase first, then a PROPOSE APPROACHES phase limited to 2–3 options, and finally a CONFIRM and STORE phase that writes confirmed outcomes into a .progress.md file.

Yes, you can enforce option count limits of 2–4 choices with a recommended-option format, allowing the interview to present focused decisions and detect completion signals to stop early when the user indicates readiness.

Parallel agent dispatch fans out multiple explore and research agents in a single step before the interview begins, preventing the anti-pattern of asking questions before discovering relevant codebase, prior memory, and current-docs facts.

You should not use this approach when the exploration space does not depend on user preferences or constraints, or when no codebase, prior memory, or library docs exist to ground the pre-discovery phase before the decision interview.