eval-reviewlisted

<purpose> Retroactive audit of an implemented AI phase's evaluation coverage. Detects whether the evaluation strategy planned in AI-SPEC.md was actually implemented. Works standalone on any MaxVision-managed AI phase. Produces a scored EVAL-REVIEW.md with gap analysis and remediation guidance. Mirrors the pattern of /maxvision:validate-phase. </purpose> <available_agent_types> Valid MaxVision subagent types (use exact names): - maxvision-eval-auditor — Audits AI eval coverage; produces scored EVAL-REVIEW.md with verdict, score/100, and critical gap count </available_agent_types> <process> ## 0. Initialize ```bash set -euo pipefail INIT=$(maxvision-sdk query init.phase-op "${PHASE_ARG:-$ARGUMENTS}") if [[ "$INIT" == @file:* ]]; then INIT=$(cat "${INIT#@file:}"); fi AUDITOR_MODEL=$(maxvision-sdk query resolve-model maxvision-eval-auditor --raw 2>/dev/null || echo "") # Text mode — non-Claude runtimes (Codex, Gemini CLI) lack AskUserQuestion. # Activates when `--text` flag is in $ARGUMENTS OR `text_mode: true` in init JSON. TEXT_MODE=false if echo "${ARGUMENTS:-}" | grep -qE '(^|[[:space:]])--text([[:space:]]|$)'; then TEXT_MODE=true; fi if [ "${TEXT_MODE_FROM_INIT:-false}" = "true" ]; then TEXT_MODE=true; fi ``` Parse JSON for: `phase_dir`, `phase_number`, `phase_name`, `phase_slug`, `padded_phase`, `phase_found`, `commit_docs`, `text_mode`. > **Text-mode handling:** when `TEXT_MODE=true`, every `AskUserQuestion` block in this skill MUST be replaced with a plain-text num