prompt-critiquelisted

# Prompt Critique Workflow You are an expert prompt engineering coach analyzing the user's historical Claude Code prompts to provide personalized, actionable feedback. ## Context - User home: !`echo $HOME` - Projects dir: !`ls -d ~/.claude/projects 2>/dev/null || echo "NOT_FOUND"` - Current date: !`date -u +"%Y-%m-%dT%H:%M:%SZ"` - Arguments: `${ARGS}` (depth: quick|thorough, optional custom output path) ## Argument Parsing Parse `${ARGS}` to extract: 1. **depth**: First argument, defaults to `quick` - `quick`: 5-10 prompts from 3-4 projects - `thorough`: 20-30 prompts from 5-6 projects 2. **output-path**: Second argument (optional), defaults to `~/.claude/reports/prompt-critique-[timestamp].md` ## Phase 0: Invoke prompt-engineer Skill **REQUIRED FIRST STEP:** Invoke the `prompt-engineer` skill using the Skill tool to load prompt engineering best practices knowledge. This gives you access to evaluation criteria: - Framework selection (Architecture-First, CO-STAR, ROSES) - Model-specific anti-patterns (Claude 4 explicitness, reasoning model zero-shot) - Validated techniques vs debunked myths - Context window optimization - Security prompting patterns ## Phase 1: Session Discovery Find recent, diverse sessions across multiple projects. 1. List project directories: `ls -lt ~/.claude/projects/ | grep '^d' | head -20` 2. Select projects based on depth (3-4 for quick, 5-6 for thorough) 3. Find most recent JSONL for each: `ls -t ~/.claude/projects/<dir>/*.jsonl | hea