score

Score Prompt

Score the provided prompt text for context engineering quality using the ContextStellar scoring API.

What to do

1. Take the prompt text from $ARGUMENTS. If no arguments provided, ask the user to paste or describe the prompt they want scored.

2. Call the ContextStellar hooks endpoint to score the prompt:

curl -s -X POST "${CONTEXTSTELLAR_BASE_URL:-https://contextstellar.com}/api/v1/hooks/claude-code" \
  -H "Authorization: ${CONTEXTSTELLAR_API_KEY}" \
  -H "Content-Type: application/json" \
  -d "{
    \"sessionId\": \"manual-score-$(date +%s)\",
    \"type\": \"PreToolUse\",
    \"toolName\": \"ManualScore\",
    \"toolInput\": { \"content\": $(echo "$PROMPT_TEXT" | jq -Rs .) }
  }"

3. Present the results clearly:

   • Overall Score and Grade (A+ through F)
   • Dimension Breakdown — show each of the 5 dimensions with their individual scores
   • Tips — list the actionable optimization suggestions
   • Token Count and potential Token Savings

4. If the score is below B (< 70), proactively suggest specific improvements the user can make.

Scoring Dimensions

Dimension	What It Measures
Token Utilization (25%)	Waste: politeness, filler, intensifiers, redundancy
Structural Clarity (23%)	XML tags, lists, headers, code blocks, imperative keywords
Specificity (20%)	Numbers, format keywords, role assignment, examples, constraints
Content Density (17%)	Lexical diversity, n-gram repetition
Cache-Friendliness (15%)	Timestamps, UUIDs, date refs, volatile data positioning

Grade Scale

• A+ (90+): Excellent — production-ready prompt
• A (80-89): Great — minor optimizations possible
• B (70-79): Good — some waste patterns detected
• C (60-69): Fair — significant optimization opportunity
• D (50-59): Poor — major restructuring recommended
• F (<50): Failing — fundamental issues