prompt-evaluator

> **AI-consumed reference.** Optimized for Claude to read during execution. > Human-readable explanation: see [docs/architecture/HIERARCHICAL_PLANNING.md](../../../docs/architecture/HIERARCHICAL_PLANNING.md) > or [docs/getting-started/](../../../docs/getting-started/) depending on topic. # Prompt Evaluator Two modes: **Usage Analytics** (how you use Claude Code) and **Prompt Quality** (evaluate/optimize a specific prompt). **For task-intake validation (before executing a user request), use the 6-dimension benchmark in `rules/core/prompt-validation.md`** — that's a different check (task completeness, not prompt craftsmanship) and should run at `/run` pre-execution per `rules/core/no-assumption.md`. --- ## Mode 1: Usage Analytics **Trigger:** `/prompts:evaluate [--days N]` ```bash node "${CLAUDE_PLUGIN_ROOT}/scripts/metrics/evaluate-prompts.cjs" --days 7 ``` Analyzes prompt logs from `.claude/metrics/prompts/{date}.jsonl`. Reports: intent distribution, feature utilization, complexity profile, suggestions, usage score (0-100). If no data: prompt-logger hook collects automatically. Return after a few sessions. --- ## Mode 2: Prompt Quality Evaluation **Trigger:** "evaluate this prompt", "optimize this prompt", or user pastes a prompt for review. ### Process **Step 1 — Classify.** Infer intent, task type (coding/creative/RAG/agent/reasoning), constraints (format, tone, tools), target model. **Step 2 — Score.** Rate 0-10 on 5 dimensions: ```toon dimensions[5]{dimen...