recipe-diagnoselisted

# Session Doctor Diagnose one Claude Code or Codex session through three independent investigations: 1. **User utterances**: evaluate how every user message requested, delegated, changed, corrected, authorized, or closed work. 2. **Work management**: evaluate how the user and agent planned, assigned, tracked, verified, and completed the work, including subagent operation. 3. **Instruction environment**: evaluate the skills, repository instructions, and attributable persistent instructions defined by the investigator, including utilization and combined contracts. The parent acts as the diagnostic decision-maker and verifier. Fresh ordinary subagents apply one self-contained axis rubric each. This separation keeps an early finding in one axis from narrowing evidence collection in the other axes. The parent establishes each investigation boundary, treats child conclusions as evidence, accepts, returns, or discards that evidence, determines causal relationships and responsible corrections, and writes an actionable improvement report in the conversation. ## Inputs Accept a session ID, transcript path, or hints that identify one session. When no selector is supplied, find transcripts associated with the current repository and order them by last recorded activity. Present the most recent candidate's harness, session ID or path, repository, time range when available, initial request, and observed endpoint for confirmation. An exact ID or path permits analysis after presenting th