mcp-rubber-duck 1.2.5 → 1.4.0

package/dist/tools/duck-judge.js

@@ -0,0 +1,208 @@
+import { logger } from '../utils/logger.js';
+const DEFAULT_CRITERIA = ['accuracy', 'completeness', 'clarity'];
+export async function duckJudgeTool(providerManager, args) {
+    const { responses, judge, criteria = DEFAULT_CRITERIA, persona, } = args;
+    // Validate inputs
+    if (!responses || !Array.isArray(responses) || responses.length === 0) {
+        throw new Error('At least one response is required to judge');
+    }
+    if (responses.length === 1) {
+        throw new Error('At least two responses are required for comparison');
+    }
+    // Determine judge provider
+    const judgeProvider = judge || providerManager.getProviderNames()[0];
+    if (!judgeProvider) {
+        throw new Error('No judge provider available');
+    }
+    logger.info(`Starting judgment with ${judgeProvider} on ${responses.length} responses`);
+    // Build the judgment prompt
+    const prompt = buildJudgePrompt(responses, criteria, persona);
+    // Get judgment from the judge duck
+    const judgeResponse = await providerManager.askDuck(judgeProvider, prompt);
+    // Parse the judgment
+    const evaluation = parseJudgment(judgeResponse.content, judgeResponse.provider, judgeResponse.nickname, responses, criteria);
+    // Format output
+    const formattedOutput = formatJudgeResult(evaluation);
+    logger.info(`Judgment completed by ${judgeProvider}: #1 is ${evaluation.rankings[0]?.provider || 'unknown'}`);
+    return {
+        content: [
+            {
+                type: 'text',
+                text: formattedOutput,
+            },
+        ],
+    };
+}
+function buildJudgePrompt(responses, criteria, persona) {
+    const criteriaList = criteria.map((c, i) => `${i + 1}. ${c}`).join('\n');
+    const responsesText = responses.map((r, i) => `--- Response ${i + 1} (${r.nickname} / ${r.provider}) ---\n${r.content}\n`).join('\n');
+    const personaText = persona
+        ? `You are a ${persona} evaluating these responses.\n\n`
+        : '';
+    return `${personaText}You are a judge evaluating ${responses.length} responses to the same prompt.
+
+RESPONSES TO EVALUATE:
+${responsesText}
+
+EVALUATION CRITERIA:
+${criteriaList}
+
+INSTRUCTIONS:
+1. Evaluate each response against ALL criteria
+2. Assign a score from 0-100 for each response
+3. Rank responses from best to worst
+4. Provide a brief justification for each ranking
+5. Give a final summary
+
+Respond with ONLY a JSON object in this exact format:
+{
+  "rankings": [
+    {"provider": "<provider name>", "score": <0-100>, "justification": "<brief explanation>"},
+    {"provider": "<provider name>", "score": <0-100>, "justification": "<brief explanation>"}
+  ],
+  "criteria_scores": {
+    "<provider>": {${criteria.map(c => `"${c}": <0-100>`).join(', ')}}
+  },
+  "summary": "<overall assessment and recommendation>"
+}
+
+IMPORTANT:
+- Rankings must be ordered from highest score to lowest
+- Use the exact provider names from the responses
+- Do NOT include any text before or after the JSON
+- Do NOT use markdown code blocks`;
+}
+function matchProvider(judgeProviderName, originalResponses) {
+    const nameLower = judgeProviderName.toLowerCase();
+    // Try exact match first
+    const exactMatch = originalResponses.find(r => r.provider.toLowerCase() === nameLower);
+    if (exactMatch)
+        return exactMatch;
+    // Try matching by provider name contained in judge's response
+    const containsMatch = originalResponses.find(r => nameLower.includes(r.provider.toLowerCase()) ||
+        nameLower.includes(r.nickname.toLowerCase()));
+    if (containsMatch)
+        return containsMatch;
+    // Try matching by nickname
+    const nicknameMatch = originalResponses.find(r => r.nickname.toLowerCase() === nameLower);
+    if (nicknameMatch)
+        return nicknameMatch;
+    return undefined;
+}
+function parseJudgment(response, judgeProvider, judgeNickname, originalResponses, criteria) {
+    const evaluation = {
+        judge: judgeProvider,
+        judgeNickname: judgeNickname,
+        prompt: '', // Will be filled by caller if needed
+        criteria,
+        rankings: [],
+        criteriaScores: {},
+        summary: '',
+        rawResponse: response,
+    };
+    try {
+        // Try to extract JSON from the response
+        const jsonMatch = response.match(/\{[\s\S]*\}/);
+        if (!jsonMatch) {
+            logger.warn(`No JSON found in judge response from ${judgeProvider}`);
+            return createFallbackEvaluation(evaluation, originalResponses, response);
+        }
+        const parsed = JSON.parse(jsonMatch[0]);
+        const matchedProviders = new Set();
+        // Parse rankings
+        if (Array.isArray(parsed.rankings)) {
+            for (const [index, r] of parsed.rankings.entries()) {
+                const matched = matchProvider(r.provider, originalResponses);
+                if (matched && !matchedProviders.has(matched.provider)) {
+                    matchedProviders.add(matched.provider);
+                    evaluation.rankings.push({
+                        provider: matched.provider,
+                        nickname: matched.nickname,
+                        rank: index + 1,
+                        score: typeof r.score === 'number' ? Math.max(0, Math.min(100, r.score)) : 0,
+                        justification: r.justification?.toString() || '',
+                    });
+                }
+            }
+        }
+        // Parse criteria scores
+        if (parsed.criteria_scores && typeof parsed.criteria_scores === 'object') {
+            evaluation.criteriaScores = parsed.criteria_scores;
+        }
+        // Parse summary
+        if (parsed.summary) {
+            evaluation.summary = parsed.summary.toString();
+        }
+    }
+    catch (error) {
+        logger.warn(`Failed to parse JSON judgment from ${judgeProvider}:`, error);
+        return createFallbackEvaluation(evaluation, originalResponses, response);
+    }
+    // Ensure all original responses are represented
+    const rankedProviders = new Set(evaluation.rankings.map(r => r.provider));
+    for (const resp of originalResponses) {
+        if (!rankedProviders.has(resp.provider)) {
+            evaluation.rankings.push({
+                provider: resp.provider,
+                nickname: resp.nickname,
+                rank: evaluation.rankings.length + 1,
+                score: 0,
+                justification: 'Not evaluated by judge',
+            });
+        }
+    }
+    return evaluation;
+}
+function createFallbackEvaluation(evaluation, originalResponses, rawResponse) {
+    // Create a basic evaluation when parsing fails
+    evaluation.rankings = originalResponses.map((r, index) => ({
+        provider: r.provider,
+        nickname: r.nickname,
+        rank: index + 1,
+        score: 50,
+        justification: 'Unable to parse judge response',
+    }));
+    evaluation.summary = `Judge evaluation parsing failed. Raw response available for review.`;
+    evaluation.rawResponse = rawResponse;
+    return evaluation;
+}
+function formatJudgeResult(evaluation) {
+    let output = `⚖️ **Judge Evaluation**\n`;
+    output += `═══════════════════════════════════════\n\n`;
+    output += `**Judge:** ${evaluation.judgeNickname} (${evaluation.judge})\n`;
+    output += `**Criteria:** ${evaluation.criteria.join(', ')}\n\n`;
+    // Rankings
+    output += `**Rankings:**\n`;
+    output += `─────────────────────────────────────\n`;
+    for (const ranking of evaluation.rankings) {
+        const medal = ranking.rank === 1 ? '🥇' : ranking.rank === 2 ? '🥈' : ranking.rank === 3 ? '🥉' : '  ';
+        const bar = '█'.repeat(Math.floor(ranking.score / 10));
+        const emptyBar = '░'.repeat(10 - Math.floor(ranking.score / 10));
+        output += `${medal} **#${ranking.rank} ${ranking.nickname}** (${ranking.provider})\n`;
+        output += `   Score: ${bar}${emptyBar} ${ranking.score}/100\n`;
+        output += `   💭 "${ranking.justification}"\n\n`;
+    }
+    // Criteria breakdown if available
+    if (Object.keys(evaluation.criteriaScores).length > 0) {
+        output += `**Criteria Breakdown:**\n`;
+        output += `─────────────────────────────────────\n`;
+        for (const [provider, scores] of Object.entries(evaluation.criteriaScores)) {
+            output += `📊 **${provider}:**\n`;
+            for (const [criterion, score] of Object.entries(scores)) {
+                const criterionScore = typeof score === 'number' ? score : 0;
+                output += `   • ${criterion}: ${criterionScore}/100\n`;
+            }
+            output += `\n`;
+        }
+    }
+    // Summary
+    if (evaluation.summary) {
+        output += `**Summary:**\n`;
+        output += `─────────────────────────────────────\n`;
+        output += `${evaluation.summary}\n\n`;
+    }
+    output += `═══════════════════════════════════════\n`;
+    output += `📋 Evaluated ${evaluation.rankings.length} responses\n`;
+    return output;
+}
+//# sourceMappingURL=duck-judge.js.map

package/dist/tools/duck-judge.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"duck-judge.js","sourceRoot":"","sources":["../../src/tools/duck-judge.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,MAAM,EAAE,MAAM,oBAAoB,CAAC;AAmB5C,MAAM,gBAAgB,GAAG,CAAC,UAAU,EAAE,cAAc,EAAE,SAAS,CAAC,CAAC;AAEjE,MAAM,CAAC,KAAK,UAAU,aAAa,CACjC,eAAgC,EAChC,IAA6B;IAE7B,MAAM,EACJ,SAAS,EACT,KAAK,EACL,QAAQ,GAAG,gBAAgB,EAC3B,OAAO,GACR,GAAG,IAAgC,CAAC;IAErC,kBAAkB;IAClB,IAAI,CAAC,SAAS,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,SAAS,CAAC,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACtE,MAAM,IAAI,KAAK,CAAC,4CAA4C,CAAC,CAAC;IAChE,CAAC;IAED,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC3B,MAAM,IAAI,KAAK,CAAC,oDAAoD,CAAC,CAAC;IACxE,CAAC;IAED,2BAA2B;IAC3B,MAAM,aAAa,GAAG,KAAK,IAAI,eAAe,CAAC,gBAAgB,EAAE,CAAC,CAAC,CAAC,CAAC;IACrE,IAAI,CAAC,aAAa,EAAE,CAAC;QACnB,MAAM,IAAI,KAAK,CAAC,6BAA6B,CAAC,CAAC;IACjD,CAAC;IAED,MAAM,CAAC,IAAI,CAAC,0BAA0B,aAAa,OAAO,SAAS,CAAC,MAAM,YAAY,CAAC,CAAC;IAExF,4BAA4B;IAC5B,MAAM,MAAM,GAAG,gBAAgB,CAAC,SAAS,EAAE,QAAQ,EAAE,OAAO,CAAC,CAAC;IAE9D,mCAAmC;IACnC,MAAM,aAAa,GAAG,MAAM,eAAe,CAAC,OAAO,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC;IAE3E,qBAAqB;IACrB,MAAM,UAAU,GAAG,aAAa,CAC9B,aAAa,CAAC,OAAO,EACrB,aAAa,CAAC,QAAQ,EACtB,aAAa,CAAC,QAAQ,EACtB,SAAS,EACT,QAAQ,CACT,CAAC;IAEF,gBAAgB;IAChB,MAAM,eAAe,GAAG,iBAAiB,CAAC,UAAU,CAAC,CAAC;IAEtD,MAAM,CAAC,IAAI,CACT,yBAAyB,aAAa,WAAW,UAAU,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,IAAI,SAAS,EAAE,CACjG,CAAC;IAEF,OAAO;QACL,OAAO,EAAE;YACP;gBACE,IAAI,EAAE,MAAM;gBACZ,IAAI,EAAE,eAAe;aACtB;SACF;KACF,CAAC;AACJ,CAAC;AAED,SAAS,gBAAgB,CACvB,SAAyB,EACzB,QAAkB,EAClB,OAAgB;IAEhB,MAAM,YAAY,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAEzE,MAAM,aAAa,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAC3C,gBAAgB,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,QAAQ,MAAM,CAAC,CAAC,QAAQ,UAAU,CAAC,CAAC,OAAO,IAAI,CAC5E,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAEb,MAAM,WAAW,GAAG,OAAO;QACzB,CAAC,CAAC,aAAa,OAAO,kCAAkC;QACxD,CAAC,CAAC,EAAE,CAAC;IAEP,OAAO,GAAG,WAAW,8BAA8B,SAAS,CAAC,MAAM;;;EAGnE,aAAa;;;EAGb,YAAY;;;;;;;;;;;;;;;;qBAgBO,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC;;;;;;;;;kCASlC,CAAC;AACnC,CAAC;AAED,SAAS,aAAa,CACpB,iBAAyB,EACzB,iBAAiC;IAEjC,MAAM,SAAS,GAAG,iBAAiB,CAAC,WAAW,EAAE,CAAC;IAElD,wBAAwB;IACxB,MAAM,UAAU,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,WAAW,EAAE,KAAK,SAAS,CAAC,CAAC;IACvF,IAAI,UAAU;QAAE,OAAO,UAAU,CAAC;IAElC,8DAA8D;IAC9D,MAAM,aAAa,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAC/C,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC;QAC5C,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC,CAC7C,CAAC;IACF,IAAI,aAAa;QAAE,OAAO,aAAa,CAAC;IAExC,2BAA2B;IAC3B,MAAM,aAAa,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAC/C,CAAC,CAAC,QAAQ,CAAC,WAAW,EAAE,KAAK,SAAS,CACvC,CAAC;IACF,IAAI,aAAa;QAAE,OAAO,aAAa,CAAC;IAExC,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,SAAS,aAAa,CACpB,QAAgB,EAChB,aAAqB,EACrB,aAAqB,EACrB,iBAAiC,EACjC,QAAkB;IAElB,MAAM,UAAU,GAAoB;QAClC,KAAK,EAAE,aAAa;QACpB,aAAa,EAAE,aAAa;QAC5B,MAAM,EAAE,EAAE,EAAE,qCAAqC;QACjD,QAAQ;QACR,QAAQ,EAAE,EAAE;QACZ,cAAc,EAAE,EAAE;QAClB,OAAO,EAAE,EAAE;QACX,WAAW,EAAE,QAAQ;KACtB,CAAC;IAEF,IAAI,CAAC;QACH,wCAAwC;QACxC,MAAM,SAAS,GAAG,QAAQ,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;QAChD,IAAI,CAAC,SAAS,EAAE,CAAC;YACf,MAAM,CAAC,IAAI,CAAC,wCAAwC,aAAa,EAAE,CAAC,CAAC;YACrE,OAAO,wBAAwB,CAAC,UAAU,EAAE,iBAAiB,EAAE,QAAQ,CAAC,CAAC;QAC3E,CAAC;QAED,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAmB,CAAC;QAC1D,MAAM,gBAAgB,GAAG,IAAI,GAAG,EAAU,CAAC;QAE3C,iBAAiB;QACjB,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;YACnC,KAAK,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC,IAAI,MAAM,CAAC,QAAQ,CAAC,OAAO,EAAE,EAAE,CAAC;gBACnD,MAAM,OAAO,GAAG,aAAa,CAAC,CAAC,CAAC,QAAQ,EAAE,iBAAiB,CAAC,CAAC;gBAC7D,IAAI,OAAO,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC;oBACvD,gBAAgB,CAAC,GAAG,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;oBACvC,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC;wBACvB,QAAQ,EAAE,OAAO,CAAC,QAAQ;wBAC1B,QAAQ,EAAE,OAAO,CAAC,QAAQ;wBAC1B,IAAI,EAAE,KAAK,GAAG,CAAC;wBACf,KAAK,EAAE,OAAO,CAAC,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;wBAC5E,aAAa,EAAE,CAAC,CAAC,aAAa,EAAE,QAAQ,EAAE,IAAI,EAAE;qBACjD,CAAC,CAAC;gBACL,CAAC;YACH,CAAC;QACH,CAAC;QAED,wBAAwB;QACxB,IAAI,MAAM,CAAC,eAAe,IAAI,OAAO,MAAM,CAAC,eAAe,KAAK,QAAQ,EAAE,CAAC;YACzE,UAAU,CAAC,cAAc,GAAG,MAAM,CAAC,eAAe,CAAC;QACrD,CAAC;QAED,gBAAgB;QAChB,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;YACnB,UAAU,CAAC,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC,QAAQ,EAAE,CAAC;QACjD,CAAC;IAEH,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,MAAM,CAAC,IAAI,CAAC,sCAAsC,aAAa,GAAG,EAAE,KAAK,CAAC,CAAC;QAC3E,OAAO,wBAAwB,CAAC,UAAU,EAAE,iBAAiB,EAAE,QAAQ,CAAC,CAAC;IAC3E,CAAC;IAED,gDAAgD;IAChD,MAAM,eAAe,GAAG,IAAI,GAAG,CAAC,UAAU,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC;IAC1E,KAAK,MAAM,IAAI,IAAI,iBAAiB,EAAE,CAAC;QACrC,IAAI,CAAC,eAAe,CAAC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,CAAC;YACxC,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC;gBACvB,QAAQ,EAAE,IAAI,CAAC,QAAQ;gBACvB,QAAQ,EAAE,IAAI,CAAC,QAAQ;gBACvB,IAAI,EAAE,UAAU,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC;gBACpC,KAAK,EAAE,CAAC;gBACR,aAAa,EAAE,wBAAwB;aACxC,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,OAAO,UAAU,CAAC;AACpB,CAAC;AAED,SAAS,wBAAwB,CAC/B,UAA2B,EAC3B,iBAAiC,EACjC,WAAmB;IAEnB,+CAA+C;IAC/C,UAAU,CAAC,QAAQ,GAAG,iBAAiB,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC;QACzD,QAAQ,EAAE,CAAC,CAAC,QAAQ;QACpB,QAAQ,EAAE,CAAC,CAAC,QAAQ;QACpB,IAAI,EAAE,KAAK,GAAG,CAAC;QACf,KAAK,EAAE,EAAE;QACT,aAAa,EAAE,gCAAgC;KAChD,CAAC,CAAC,CAAC;IACJ,UAAU,CAAC,OAAO,GAAG,qEAAqE,CAAC;IAC3F,UAAU,CAAC,WAAW,GAAG,WAAW,CAAC;IACrC,OAAO,UAAU,CAAC;AACpB,CAAC;AAED,SAAS,iBAAiB,CAAC,UAA2B;IACpD,IAAI,MAAM,GAAG,2BAA2B,CAAC;IACzC,MAAM,IAAI,6CAA6C,CAAC;IACxD,MAAM,IAAI,cAAc,UAAU,CAAC,aAAa,KAAK,UAAU,CAAC,KAAK,KAAK,CAAC;IAC3E,MAAM,IAAI,iBAAiB,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC;IAEhE,WAAW;IACX,MAAM,IAAI,iBAAiB,CAAC;IAC5B,MAAM,IAAI,yCAAyC,CAAC;IAEpD,KAAK,MAAM,OAAO,IAAI,UAAU,CAAC,QAAQ,EAAE,CAAC;QAC1C,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC;QACvG,MAAM,GAAG,GAAG,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,GAAG,EAAE,CAAC,CAAC,CAAC;QACvD,MAAM,QAAQ,GAAG,GAAG,CAAC,MAAM,CAAC,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,GAAG,EAAE,CAAC,CAAC,CAAC;QAEjE,MAAM,IAAI,GAAG,KAAK,OAAO,OAAO,CAAC,IAAI,IAAI,OAAO,CAAC,QAAQ,OAAO,OAAO,CAAC,QAAQ,KAAK,CAAC;QACtF,MAAM,IAAI,aAAa,GAAG,GAAG,QAAQ,IAAI,OAAO,CAAC,KAAK,QAAQ,CAAC;QAC/D,MAAM,IAAI,UAAU,OAAO,CAAC,aAAa,OAAO,CAAC;IACnD,CAAC;IAED,kCAAkC;IAClC,IAAI,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC,cAAc,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACtD,MAAM,IAAI,2BAA2B,CAAC;QACtC,MAAM,IAAI,yCAAyC,CAAC;QAEpD,KAAK,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,cAAc,CAAC,EAAE,CAAC;YAC3E,MAAM,IAAI,QAAQ,QAAQ,OAAO,CAAC;YAClC,KAAK,MAAM,CAAC,SAAS,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;gBACxD,MAAM,cAAc,GAAG,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;gBAC7D,MAAM,IAAI,QAAQ,SAAS,KAAK,cAAc,QAAQ,CAAC;YACzD,CAAC;YACD,MAAM,IAAI,IAAI,CAAC;QACjB,CAAC;IACH,CAAC;IAED,UAAU;IACV,IAAI,UAAU,CAAC,OAAO,EAAE,CAAC;QACvB,MAAM,IAAI,gBAAgB,CAAC;QAC3B,MAAM,IAAI,yCAAyC,CAAC;QACpD,MAAM,IAAI,GAAG,UAAU,CAAC,OAAO,MAAM,CAAC;IACxC,CAAC;IAED,MAAM,IAAI,2CAA2C,CAAC;IACtD,MAAM,IAAI,gBAAgB,UAAU,CAAC,QAAQ,CAAC,MAAM,cAAc,CAAC;IAEnE,OAAO,MAAM,CAAC;AAChB,CAAC"}

package/dist/tools/duck-vote.d.ts

@@ -0,0 +1,14 @@
+import { ProviderManager } from '../providers/manager.js';
+export interface DuckVoteArgs {
+    question: string;
+    options: string[];
+    voters?: string[];
+    require_reasoning?: boolean;
+}
+export declare function duckVoteTool(providerManager: ProviderManager, args: Record<string, unknown>): Promise<{
+    content: {
+        type: string;
+        text: string;
+    }[];
+}>;
+//# sourceMappingURL=duck-vote.d.ts.map

package/dist/tools/duck-vote.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"duck-vote.d.ts","sourceRoot":"","sources":["../../src/tools/duck-vote.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,yBAAyB,CAAC;AAK1D,MAAM,WAAW,YAAY;IAC3B,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;IAClB,iBAAiB,CAAC,EAAE,OAAO,CAAC;CAC7B;AAED,wBAAsB,YAAY,CAChC,eAAe,EAAE,eAAe,EAChC,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC;;;;;GAwE9B"}

package/dist/tools/duck-vote.js

@@ -0,0 +1,46 @@
+import { ConsensusService } from '../services/consensus.js';
+import { logger } from '../utils/logger.js';
+export async function duckVoteTool(providerManager, args) {
+    const { question, options, voters, require_reasoning = true, } = args;
+    // Validate inputs
+    if (!question || typeof question !== 'string') {
+        throw new Error('Question is required');
+    }
+    if (!options || !Array.isArray(options) || options.length < 2) {
+        throw new Error('At least 2 options are required');
+    }
+    if (options.length > 10) {
+        throw new Error('Maximum 10 options allowed');
+    }
+    // Get voters (all providers if not specified)
+    const voterNames = voters && voters.length > 0
+        ? voters
+        : providerManager.getProviderNames();
+    if (voterNames.length === 0) {
+        throw new Error('No voters available');
+    }
+    logger.info(`Starting vote with ${voterNames.length} voters on: "${question}"`);
+    const consensusService = new ConsensusService();
+    const votePrompt = consensusService.buildVotePrompt(question, options, require_reasoning);
+    // Get votes from all ducks in parallel
+    const responses = await providerManager.compareDucks(votePrompt, voterNames);
+    // Parse votes
+    const votes = responses.map(response => {
+        return consensusService.parseVote(response.content, response.provider, response.nickname, options);
+    });
+    // Aggregate results
+    const aggregatedResult = consensusService.aggregateVotes(question, options, votes);
+    // Format output
+    const formattedOutput = consensusService.formatVoteResult(aggregatedResult);
+    logger.info(`Vote completed: ${aggregatedResult.consensusLevel} consensus, ` +
+        `winner: ${aggregatedResult.winner || 'none'}`);
+    return {
+        content: [
+            {
+                type: 'text',
+                text: formattedOutput,
+            },
+        ],
+    };
+}
+//# sourceMappingURL=duck-vote.js.map

package/dist/tools/duck-vote.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"duck-vote.js","sourceRoot":"","sources":["../../src/tools/duck-vote.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,gBAAgB,EAAE,MAAM,0BAA0B,CAAC;AAE5D,OAAO,EAAE,MAAM,EAAE,MAAM,oBAAoB,CAAC;AAS5C,MAAM,CAAC,KAAK,UAAU,YAAY,CAChC,eAAgC,EAChC,IAA6B;IAE7B,MAAM,EACJ,QAAQ,EACR,OAAO,EACP,MAAM,EACN,iBAAiB,GAAG,IAAI,GACzB,GAAG,IAA+B,CAAC;IAEpC,kBAAkB;IAClB,IAAI,CAAC,QAAQ,IAAI,OAAO,QAAQ,KAAK,QAAQ,EAAE,CAAC;QAC9C,MAAM,IAAI,KAAK,CAAC,sBAAsB,CAAC,CAAC;IAC1C,CAAC;IAED,IAAI,CAAC,OAAO,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC9D,MAAM,IAAI,KAAK,CAAC,iCAAiC,CAAC,CAAC;IACrD,CAAC;IAED,IAAI,OAAO,CAAC,MAAM,GAAG,EAAE,EAAE,CAAC;QACxB,MAAM,IAAI,KAAK,CAAC,4BAA4B,CAAC,CAAC;IAChD,CAAC;IAED,8CAA8C;IAC9C,MAAM,UAAU,GAAG,MAAM,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC;QAC5C,CAAC,CAAC,MAAM;QACR,CAAC,CAAC,eAAe,CAAC,gBAAgB,EAAE,CAAC;IAEvC,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC5B,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;IACzC,CAAC;IAED,MAAM,CAAC,IAAI,CAAC,sBAAsB,UAAU,CAAC,MAAM,gBAAgB,QAAQ,GAAG,CAAC,CAAC;IAEhF,MAAM,gBAAgB,GAAG,IAAI,gBAAgB,EAAE,CAAC;IAChD,MAAM,UAAU,GAAG,gBAAgB,CAAC,eAAe,CACjD,QAAQ,EACR,OAAO,EACP,iBAAiB,CAClB,CAAC;IAEF,uCAAuC;IACvC,MAAM,SAAS,GAAG,MAAM,eAAe,CAAC,YAAY,CAAC,UAAU,EAAE,UAAU,CAAC,CAAC;IAE7E,cAAc;IACd,MAAM,KAAK,GAAiB,SAAS,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE;QACnD,OAAO,gBAAgB,CAAC,SAAS,CAC/B,QAAQ,CAAC,OAAO,EAChB,QAAQ,CAAC,QAAQ,EACjB,QAAQ,CAAC,QAAQ,EACjB,OAAO,CACR,CAAC;IACJ,CAAC,CAAC,CAAC;IAEH,oBAAoB;IACpB,MAAM,gBAAgB,GAAG,gBAAgB,CAAC,cAAc,CAAC,QAAQ,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC;IAEnF,gBAAgB;IAChB,MAAM,eAAe,GAAG,gBAAgB,CAAC,gBAAgB,CAAC,gBAAgB,CAAC,CAAC;IAE5E,MAAM,CAAC,IAAI,CACT,mBAAmB,gBAAgB,CAAC,cAAc,cAAc;QAChE,WAAW,gBAAgB,CAAC,MAAM,IAAI,MAAM,EAAE,CAC/C,CAAC;IAEF,OAAO;QACL,OAAO,EAAE;YACP;gBACE,IAAI,EAAE,MAAM;gBACZ,IAAI,EAAE,eAAe;aACtB;SACF;KACF,CAAC;AACJ,CAAC"}

package/docker-compose.yml

@@ -48,7 +48,7 @@ services:
       
       # OpenAI Provider
       - OPENAI_API_KEY=${OPENAI_API_KEY}
-      - OPENAI_DEFAULT_MODEL=${OPENAI_DEFAULT_MODEL:-gpt-4o-mini}
+      - OPENAI_DEFAULT_MODEL=${OPENAI_DEFAULT_MODEL:-gpt-5.1}
       
       # Google Gemini Provider
       - GEMINI_API_KEY=${GEMINI_API_KEY}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mcp-rubber-duck",
-  "version": "1.2.5",
+  "version": "1.4.0",
   "description": "An MCP server that bridges to multiple OpenAI-compatible LLMs - your AI rubber duck debugging panel",
   "main": "dist/index.js",
   "type": "module",

package/src/config/config.ts

@@ -109,8 +109,8 @@ export class ConfigManager {
       providers.openai = {
         api_key: process.env.OPENAI_API_KEY,
         base_url: 'https://api.openai.com/v1',
-        models: ['gpt-4o', 'gpt-4o-mini', 'gpt-3.5-turbo'],
-        default_model: process.env.OPENAI_DEFAULT_MODEL || 'gpt-4o-mini',
+        models: ['gpt-5.1', 'gpt-4.1', 'gpt-4o'],
+        default_model: process.env.OPENAI_DEFAULT_MODEL || 'gpt-5.1',
         nickname: process.env.OPENAI_NICKNAME || 'GPT Duck',
       };
     }
@@ -120,7 +120,7 @@ export class ConfigManager {
       providers.gemini = {
         api_key: process.env.GEMINI_API_KEY,
         base_url: 'https://generativelanguage.googleapis.com/v1beta/openai/',
-        models: ['gemini-2.5-flash', 'gemini-2.0-flash'],
+        models: ['gemini-3-pro-preview', 'gemini-2.5-pro', 'gemini-2.5-flash'],
         default_model: process.env.GEMINI_DEFAULT_MODEL || 'gemini-2.5-flash',
         nickname: process.env.GEMINI_NICKNAME || 'Gemini Duck',
       };
@@ -131,7 +131,7 @@ export class ConfigManager {
       providers.groq = {
         api_key: process.env.GROQ_API_KEY,
         base_url: 'https://api.groq.com/openai/v1',
-        models: ['llama-3.3-70b-versatile', 'mixtral-8x7b-32768'],
+        models: ['meta-llama/llama-4-scout-17b-16e-instruct', 'meta-llama/llama-4-maverick-17b-128e-instruct', 'llama-3.3-70b-versatile'],
         default_model: process.env.GROQ_DEFAULT_MODEL || 'llama-3.3-70b-versatile',
         nickname: process.env.GROQ_NICKNAME || 'Groq Duck',
       };

package/src/config/types.ts

@@ -86,4 +86,96 @@ export interface DuckResponse {
   };
   latency: number;
   cached: boolean;
+}
+
+// Consensus & Voting Types
+export interface VoteResult {
+  voter: string;
+  nickname: string;
+  choice: string;
+  confidence: number;
+  reasoning: string;
+  rawResponse: string;
+}
+
+export interface AggregatedVote {
+  question: string;
+  options: string[];
+  winner: string | null;
+  isTie: boolean;
+  tally: Record<string, number>;
+  confidenceByOption: Record<string, number>;
+  votes: VoteResult[];
+  totalVoters: number;
+  validVotes: number;
+  consensusLevel: 'unanimous' | 'majority' | 'plurality' | 'split' | 'none';
+}
+
+// Judge Evaluation Types
+export interface JudgeRanking {
+  provider: string;
+  nickname: string;
+  rank: number;
+  score: number;
+  justification: string;
+}
+
+export interface JudgeEvaluation {
+  judge: string;
+  judgeNickname: string;
+  prompt: string;
+  criteria: string[];
+  rankings: JudgeRanking[];
+  criteriaScores: Record<string, Record<string, number>>;
+  summary: string;
+  rawResponse: string;
+}
+
+// Iteration Types
+export interface IterationRound {
+  round: number;
+  provider: string;
+  nickname: string;
+  role: 'generator' | 'critic' | 'refiner';
+  content: string;
+  timestamp: Date;
+}
+
+export interface IterationResult {
+  prompt: string;
+  mode: 'refine' | 'critique-improve';
+  providers: [string, string];
+  rounds: IterationRound[];
+  finalResponse: string;
+  totalIterations: number;
+  converged: boolean;
+}
+
+// Debate Types
+export type DebateFormat = 'oxford' | 'socratic' | 'adversarial';
+export type DebatePosition = 'pro' | 'con' | 'neutral';
+
+export interface DebateParticipant {
+  provider: string;
+  nickname: string;
+  position: DebatePosition;
+}
+
+export interface DebateArgument {
+  round: number;
+  provider: string;
+  nickname: string;
+  position: DebatePosition;
+  content: string;
+  timestamp: Date;
+}
+
+export interface DebateResult {
+  topic: string;
+  format: DebateFormat;
+  participants: DebateParticipant[];
+  rounds: DebateArgument[][];
+  synthesis: string;
+  synthesizer: string;
+  totalRounds: number;
 }

package/src/server.ts

@@ -27,6 +27,10 @@ import { listDucksTool } from './tools/list-ducks.js';
 import { listModelsTool } from './tools/list-models.js';
 import { compareDucksTool } from './tools/compare-ducks.js';
 import { duckCouncilTool } from './tools/duck-council.js';
+import { duckVoteTool } from './tools/duck-vote.js';
+import { duckJudgeTool } from './tools/duck-judge.js';
+import { duckIterateTool } from './tools/duck-iterate.js';
+import { duckDebateTool } from './tools/duck-debate.js';
 
 // Import MCP tools
 import { getPendingApprovalsTool } from './tools/get-pending-approvals.js';
@@ -162,6 +166,18 @@ export class RubberDuckServer {
             }
             return await duckCouncilTool(this.providerManager, args || {});
 
+          case 'duck_vote':
+            return await duckVoteTool(this.providerManager, args || {});
+
+          case 'duck_judge':
+            return await duckJudgeTool(this.providerManager, args || {});
+
+          case 'duck_iterate':
+            return await duckIterateTool(this.providerManager, args || {});
+
+          case 'duck_debate':
+            return await duckDebateTool(this.providerManager, args || {});
+
           // MCP-specific tools
           case 'get_pending_approvals':
             if (!this.approvalService) {
@@ -487,6 +503,144 @@ export class RubberDuckServer {
           required: ['prompt'],
         },
       },
+      {
+        name: 'duck_vote',
+        description: 'Have multiple ducks vote on options with reasoning. Returns vote tally, confidence scores, and consensus level.',
+        inputSchema: {
+          type: 'object',
+          properties: {
+            question: {
+              type: 'string',
+              description: 'The question to vote on (e.g., "Best approach for error handling?")',
+            },
+            options: {
+              type: 'array',
+              items: { type: 'string' },
+              minItems: 2,
+              maxItems: 10,
+              description: 'The options to vote on (2-10 options)',
+            },
+            voters: {
+              type: 'array',
+              items: { type: 'string' },
+              description: 'List of provider names to vote (optional, uses all if not specified)',
+            },
+            require_reasoning: {
+              type: 'boolean',
+              default: true,
+              description: 'Require ducks to explain their vote (default: true)',
+            },
+          },
+          required: ['question', 'options'],
+        },
+      },
+      {
+        name: 'duck_judge',
+        description: 'Have one duck evaluate and rank other ducks\' responses. Use after duck_council to get a comparative evaluation.',
+        inputSchema: {
+          type: 'object',
+          properties: {
+            responses: {
+              type: 'array',
+              items: {
+                type: 'object',
+                properties: {
+                  provider: { type: 'string' },
+                  nickname: { type: 'string' },
+                  model: { type: 'string' },
+                  content: { type: 'string' },
+                },
+                required: ['provider', 'nickname', 'content'],
+              },
+              minItems: 2,
+              description: 'Array of duck responses to evaluate (from duck_council output)',
+            },
+            judge: {
+              type: 'string',
+              description: 'Provider name of the judge duck (optional, uses first available)',
+            },
+            criteria: {
+              type: 'array',
+              items: { type: 'string' },
+              description: 'Evaluation criteria (default: ["accuracy", "completeness", "clarity"])',
+            },
+            persona: {
+              type: 'string',
+              description: 'Judge persona (e.g., "senior engineer", "security expert")',
+            },
+          },
+          required: ['responses'],
+        },
+      },
+      {
+        name: 'duck_iterate',
+        description: 'Iteratively refine a response between two ducks. One generates, the other critiques/improves, alternating for multiple rounds.',
+        inputSchema: {
+          type: 'object',
+          properties: {
+            prompt: {
+              type: 'string',
+              description: 'The initial prompt/task to iterate on',
+            },
+            iterations: {
+              type: 'number',
+              minimum: 1,
+              maximum: 10,
+              default: 3,
+              description: 'Number of iteration rounds (default: 3, max: 10)',
+            },
+            providers: {
+              type: 'array',
+              items: { type: 'string' },
+              minItems: 2,
+              maxItems: 2,
+              description: 'Exactly 2 provider names for the ping-pong iteration',
+            },
+            mode: {
+              type: 'string',
+              enum: ['refine', 'critique-improve'],
+              description: 'refine: each duck improves the previous response. critique-improve: alternates between critiquing and improving.',
+            },
+          },
+          required: ['prompt', 'providers', 'mode'],
+        },
+      },
+      {
+        name: 'duck_debate',
+        description: 'Structured multi-round debate between ducks. Supports oxford (pro/con), socratic (questioning), and adversarial (attack/defend) formats.',
+        inputSchema: {
+          type: 'object',
+          properties: {
+            prompt: {
+              type: 'string',
+              description: 'The debate topic or proposition',
+            },
+            rounds: {
+              type: 'number',
+              minimum: 1,
+              maximum: 10,
+              default: 3,
+              description: 'Number of debate rounds (default: 3)',
+            },
+            providers: {
+              type: 'array',
+              items: { type: 'string' },
+              minItems: 2,
+              description: 'Provider names to participate (min 2, uses all if not specified)',
+            },
+            format: {
+              type: 'string',
+              enum: ['oxford', 'socratic', 'adversarial'],
+              description: 'Debate format: oxford (pro/con), socratic (questioning), adversarial (attack/defend)',
+            },
+            synthesizer: {
+              type: 'string',
+              description: 'Provider to synthesize the debate (optional, uses first provider)',
+            },
+          },
+          required: ['prompt', 'format'],
+        },
+      },
     ];
 
     // Add MCP-specific tools if enabled