orbitant-debrief

Overview

A bottom-up friction-capture skill. At the end of a Claude Code session, scan the conversation already in context, identify project-specific friction (project rules and project knowledge), and propose CLAUDE.md edits the engineer can approve. Bias toward zero proposals — returning "session looks clean" is a first-class outcome, not a fallback.

This skill is the bottom-up counterpart to /orbitant-engineering:ground-control (the top-down standards audit). Together they form a feedback loop: ground-control enforces Orbitant-wide standards, debrief captures project-specific friction that may eventually graduate into a standard.

When to Use

• The engineer types /debrief (no args) at session end.
• The engineer types /debrief <hint> to focus on a specific area (e.g. /debrief the API naming convention thing).
• Skip with the message "No active session content to debrief — run me at the end of a session that had real engineering work." if the session has no engineering work in it (e.g. a brand-new chat with only this /debrief command).

Prerequisites

• The skill assumes Claude has the full Claude Code session already in context. It does not re-read the session from disk.
• The project may or may not have a CLAUDE.md. The skill handles both cases (see the application step).

Process