interview-session

The retrospective process works by identifying and replaying a closed Claude Code session to conduct a structured interview. It applies deterministic analysis to uncover failure causes and collect actionable feedback from the participant's perspective without invoking the model.

Yes, participant-perspective feedback is collected without model invocation. The skill operates deterministically with safe prompts to ensure repeatable interviews and consistent questioning across sessions, preventing non-deterministic model behavior during the retrospective.

The best way to identify failure causes is performing a structured retrospective on the closed session. By replaying the session and guiding participants through reflective questions, you can uncover root causes and gather actionable feedback to improve future performance and process changes.

To start an interview session for process improvement, invoke the skill with a specific session-id. This reopens the past Claude Code session and guides participants through reflective questions to identify failures and improvement opportunities for future tooling.

The limitation of using deterministic prompts for session interviews is the lack of dynamic model invocation. While this ensures safe and repeatable results, it restricts the retrospective to structured, predefined questioning without adaptive follow-up generation based on participant responses.