score_prompt

Implementation Reference

• src/tools/score.ts:7-16 (handler)

  The main handler function for the score_prompt tool. It applies all rule sets (efficiency, clarity, specificity, completeness) to the prompt, computes a total score and per-dimension breakdown using the scorer, and returns results including issues.

  export function scorePrompt(rawPrompt: string) {
    const rules = [
      ...applyEfficiencyRules(rawPrompt).results,
      ...applyClarityRules(rawPrompt).results,
      ...applySpecificityRules(rawPrompt).results,
      ...applyCompletenessRules(rawPrompt).results,
    ];
    const { total, breakdown } = computeScore(rules);
    return { total, breakdown, issues: rules };
  }

• src/index.ts:17-22 (registration)

  Registration of the 'score_prompt' tool on the MCP server (stdio transport). Defines the schema: expects a single 'raw_prompt' string. Calls scorePrompt() and returns JSON output.

  server.tool("score_prompt",
    { raw_prompt: z.string() },
    async ({ raw_prompt }) => ({
      content: [{ type: "text", text: JSON.stringify(scorePrompt(raw_prompt), null, 2) }]
    })
  );

• src/http.ts:22-27 (registration)

  Registration of the 'score_prompt' tool on the MCP server (HTTP/streamable transport). Same schema and handler logic as the stdio registration.

  server.tool("score_prompt",
    { raw_prompt: z.string() },
    async ({ raw_prompt }) => ({
      content: [{ type: "text", text: JSON.stringify(scorePrompt(raw_prompt), null, 2) }]
    })
  );

• src/lib/scorer.ts:13-30 (helper)

  The computeScore function that calculates the total prompt score (out of 100) and per-dimension breakdown (each up to 25 points) based on penalties derived from rule violations.

  export function computeScore(rules: RuleResult[]): { total: number; breakdown: ScoreBreakdown } {
    const penalties: Record<string, number> = { clarity: 0, specificity: 0, completeness: 0, efficiency: 0 };
  
    for (const r of rules) {
      const maxPenalty = 25 / RULES_PER_DIMENSION[r.dimension];
      penalties[r.dimension] += maxPenalty * SEVERITY_WEIGHT[r.severity];
    }
  
    const breakdown: ScoreBreakdown = {
      clarity:      Math.round(Math.max(0, 25 - penalties.clarity)),
      specificity:  Math.round(Math.max(0, 25 - penalties.specificity)),
      completeness: Math.round(Math.max(0, 25 - penalties.completeness)),
      efficiency:   Math.round(Math.max(0, 25 - penalties.efficiency)),
    };
  
    const total = (Object.values(breakdown) as number[]).reduce((a, b) => a + b, 0);
    return { total, breakdown };
  }

• src/lib/types.ts:1-17 (schema)

  Type definitions used by the scoring system: RuleResult (includes dimension, severity, message, fix_applied) and ScoreBreakdown (clarity, specificity, completeness, efficiency).

  export type Severity = "info" | "warn" | "critical";
  export type Dimension = "clarity" | "specificity" | "completeness" | "efficiency";
  
  export interface RuleResult {
    id: string;
    dimension: Dimension;
    severity: Severity;
    message: string;
    fix_applied: boolean;
  }
  
  export interface ScoreBreakdown {
    clarity: number;
    specificity: number;
    completeness: number;
    efficiency: number;
  }