evaluatorlisted

> Independent evaluation engine. 5-layer architecture, 6 scoring dimensions, anti-gaming defenses, strategic direction guidance. Works standalone or integrated with evolve-loop. ## Contents - [Quick Start](#quick-start) - [Architecture](#architecture) - [Stage 1: SCOPE](#stage-1-scope) - [Stage 2: GRADE](#stage-2-grade) - [Stage 3: DETECT](#stage-3-detect) - [Stage 4: SCORE](#stage-4-score) - [Stage 5: DIRECT](#stage-5-direct) - [Isolation Principles](#isolation-principles) - [Depth Control](#depth-control) - [Evolve-Loop Integration](#evolve-loop-integration) - [Output Schema](#output-schema) - [Reference](#reference-read-on-demand) ## Quick Start ```bash # Evaluate recent changes (task scope) /evaluator --scope task # Full project health assessment /evaluator --scope project --depth deep # Strategic direction guidance — what should improve next? /evaluator --scope strategic # Evaluate specific files /evaluator src/auth/ --scope task --depth standard ``` **Parse arguments:** - First positional → `target` (file/directory path, or omit for auto-detect from git diff) - `--scope task|project|strategic` → evaluation breadth (default: `task`) - `--depth quick|standard|deep` → analysis thoroughness (default: `standard`) ## Architecture 5-layer evaluation engine: **SCOPE → GRADE → DETECT → SCORE → DIRECT** ``` Layer 5: META-EVAL ──── Eval-of-eval: red-team the evaluator itself │ (runs periodically, not every invocation) Layer 4: DIRECT ──────── Stra