@j0hanz/code-review-analyst-mcp 1.4.2 → 1.4.4

package/dist/lib/gemini.js

@@ -4,7 +4,7 @@ import { EventEmitter } from 'node:events';
 import { performance } from 'node:perf_hooks';
 import { setTimeout as sleep } from 'node:timers/promises';
 import { debuglog } from 'node:util';
-import { GoogleGenAI, HarmBlockThreshold, HarmCategory } from '@google/genai';
+import { FinishReason, GoogleGenAI, HarmBlockThreshold, HarmCategory, } from '@google/genai';
 import { createCachedEnvInt } from './env-config.js';
 import { getErrorMessage, RETRYABLE_UPSTREAM_ERROR_PATTERN } from './errors.js';
 // Lazy-cached: first call happens after parseCommandLineArgs() sets GEMINI_MODEL.
@@ -36,8 +36,8 @@ const DIGITS_ONLY_PATTERN = /^\d+$/;
 const SLEEP_UNREF_OPTIONS = { ref: false };
 const maxConcurrentCallsConfig = createCachedEnvInt('MAX_CONCURRENT_CALLS', 10);
 const concurrencyWaitMsConfig = createCachedEnvInt('MAX_CONCURRENT_CALLS_WAIT_MS', 2_000);
-const concurrencyPollMsConfig = createCachedEnvInt('MAX_CONCURRENT_CALLS_POLL_MS', 25);
 let activeCalls = 0;
+const slotWaiters = [];
 const RETRYABLE_TRANSIENT_CODES = new Set([
     'RESOURCE_EXHAUSTED',
     'UNAVAILABLE',
@@ -370,14 +370,20 @@ async function executeAttempt(request, model, timeoutMs, attempt, onLog) {
     const startedAt = performance.now();
     const response = await generateContentWithTimeout(request, model, timeoutMs);
     const latencyMs = Math.round(performance.now() - startedAt);
+    const finishReason = response.candidates?.[0]?.finishReason;
     await emitGeminiLog(onLog, 'info', {
         event: 'gemini_call',
         details: {
             attempt,
             latencyMs,
+            finishReason: finishReason ?? null,
             usageMetadata: response.usageMetadata ?? null,
         },
     });
+    if (finishReason === FinishReason.MAX_TOKENS) {
+        const limit = request.maxOutputTokens ?? DEFAULT_MAX_OUTPUT_TOKENS;
+        throw new Error(`Response truncated: model output exceeds limit (maxOutputTokens=${formatNumber(limit)}). Increase maxOutputTokens or reduce prompt complexity.`);
+    }
     return parseStructuredResponse(response.text);
 }
 async function waitBeforeRetry(attempt, error, onLog, requestSignal) {
@@ -437,19 +443,62 @@ async function runWithRetries(request, model, timeoutMs, maxRetries, onLog) {
 function canRetryAttempt(attempt, maxRetries, error) {
     return attempt < maxRetries && shouldRetry(error);
 }
+function tryWakeNextWaiter() {
+    const next = slotWaiters.shift();
+    if (next !== undefined) {
+        next();
+    }
+}
 async function waitForConcurrencySlot(limit, requestSignal) {
+    if (activeCalls < limit) {
+        return;
+    }
+    if (requestSignal?.aborted) {
+        throw new Error('Gemini request was cancelled.');
+    }
     const waitLimitMs = concurrencyWaitMsConfig.get();
-    const pollMs = concurrencyPollMsConfig.get();
-    const deadline = performance.now() + waitLimitMs;
-    while (activeCalls >= limit) {
-        if (requestSignal?.aborted) {
-            throw new Error('Gemini request was cancelled.');
-        }
-        if (performance.now() >= deadline) {
-            throw new Error(formatConcurrencyLimitErrorMessage(limit, waitLimitMs));
+    return new Promise((resolve, reject) => {
+        let settled = false;
+        const waiter = () => {
+            if (settled)
+                return;
+            settled = true;
+            clearTimeout(deadlineTimer);
+            if (requestSignal) {
+                requestSignal.removeEventListener('abort', onAbort);
+            }
+            resolve();
+        };
+        slotWaiters.push(waiter);
+        const deadlineTimer = setTimeout(() => {
+            if (settled)
+                return;
+            settled = true;
+            const idx = slotWaiters.indexOf(waiter);
+            if (idx !== -1) {
+                slotWaiters.splice(idx, 1);
+            }
+            if (requestSignal) {
+                requestSignal.removeEventListener('abort', onAbort);
+            }
+            reject(new Error(formatConcurrencyLimitErrorMessage(limit, waitLimitMs)));
+        }, waitLimitMs);
+        deadlineTimer.unref();
+        const onAbort = () => {
+            if (settled)
+                return;
+            settled = true;
+            const idx = slotWaiters.indexOf(waiter);
+            if (idx !== -1) {
+                slotWaiters.splice(idx, 1);
+            }
+            clearTimeout(deadlineTimer);
+            reject(new Error('Gemini request was cancelled.'));
+        };
+        if (requestSignal) {
+            requestSignal.addEventListener('abort', onAbort, { once: true });
         }
-        await sleep(pollMs, undefined, SLEEP_UNREF_OPTIONS);
-    }
+    });
 }
 export async function generateStructuredJson(request) {
     const model = request.model ?? getDefaultModel();
@@ -464,5 +513,6 @@ export async function generateStructuredJson(request) {
     }
     finally {
         activeCalls -= 1;
+        tryWakeNextWaiter();
     }
 }

package/dist/lib/model-config.d.ts

@@ -2,24 +2,39 @@
 export declare const FLASH_MODEL = "gemini-2.5-flash";
 /** High-capability model for deep reasoning, quality inspection, and reliable code generation. */
 export declare const PRO_MODEL = "gemini-2.5-pro";
-/** Thinking budget (tokens) for Flash model thinking tasks (test plans, search/replace). */
-export declare const FLASH_THINKING_BUDGET: 8192;
-/** Thinking budget (tokens) for Pro model deep-analysis tasks (code quality inspection). */
-export declare const PRO_THINKING_BUDGET: 16384;
+/**
+ * Thinking budget (tokens) for Flash triage tools (impact, summary, API-breaking).
+ * Explicitly disabled (0) — these are classification/extraction tasks that do not
+ * benefit from a reasoning chain. Avoids default dynamic-thinking overhead.
+ * Flash 2.5 range: 0–24_576.
+ */
+export declare const FLASH_TRIAGE_THINKING_BUDGET: 0;
+/** Thinking budget (tokens) for Flash analysis tasks (test plans, complexity). */
+export declare const FLASH_THINKING_BUDGET: 16384;
+/** Thinking budget (tokens) for Pro model deep-analysis tasks (quality, patches). */
+export declare const PRO_THINKING_BUDGET: 24576;
 /** Output cap for Flash triage tools (impact, summary). */
-export declare const FLASH_TRIAGE_MAX_OUTPUT_TOKENS: 2048;
+export declare const FLASH_TRIAGE_MAX_OUTPUT_TOKENS: 4096;
 /** Output cap for API breaking-change detection (migration guidance needs room). */
 export declare const FLASH_API_BREAKING_MAX_OUTPUT_TOKENS: 4096;
 /** Output cap for test-plan generation (includes pseudocode snippets). */
-export declare const FLASH_TEST_PLAN_MAX_OUTPUT_TOKENS: 4096;
+export declare const FLASH_TEST_PLAN_MAX_OUTPUT_TOKENS: 8192;
 /** Output cap for Pro deep review findings. */
-export declare const PRO_REVIEW_MAX_OUTPUT_TOKENS: 8192;
+export declare const PRO_REVIEW_MAX_OUTPUT_TOKENS: 12288;
 /** Output cap for Pro search/replace remediation blocks. */
-export declare const PRO_PATCH_MAX_OUTPUT_TOKENS: 4096;
+export declare const PRO_PATCH_MAX_OUTPUT_TOKENS: 8192;
 /** Output cap for Flash complexity analysis reports. */
 export declare const FLASH_COMPLEXITY_MAX_OUTPUT_TOKENS: 2048;
 /** Extended timeout for Pro model calls (ms). Pro thinks longer than Flash. */
 export declare const DEFAULT_TIMEOUT_PRO_MS = 120000;
+/** Temperature for triage/classification tools (deterministic structured extraction). */
+export declare const TRIAGE_TEMPERATURE: 0.1;
+/** Temperature for analytical tools (consistent algorithmic reasoning). */
+export declare const ANALYSIS_TEMPERATURE: 0.1;
+/** Temperature for code patch generation (maximum precision for search blocks). */
+export declare const PATCH_TEMPERATURE: 0;
+/** Temperature for creative synthesis tools (test plan generation). */
+export declare const CREATIVE_TEMPERATURE: 0.2;
 export declare const MODEL_TIMEOUT_MS: {
     readonly defaultPro: 120000;
 };

package/dist/lib/model-config.js

@@ -3,21 +3,71 @@ export const FLASH_MODEL = 'gemini-2.5-flash';
 /** High-capability model for deep reasoning, quality inspection, and reliable code generation. */
 export const PRO_MODEL = 'gemini-2.5-pro';
 const THINKING_BUDGET_TOKENS = {
-    flash: 8_192,
-    pro: 16_384,
+    /**
+     * Disabled (0): triage/classification tasks need no reasoning chain.
+     * Flash 2.5 range: 0–24_576; 0 explicitly disables thinking.
+     */
+    flashTriage: 0,
+    /**
+     * Raised from 8_192 → half of Flash max (24_576).
+     * Used for analysis tasks (test plans, complexity) that benefit from
+     * multi-step reasoning but not from unbounded thinking tokens.
+     */
+    flash: 16_384,
+    /**
+     * Raised from 16_384 → 75 % of Pro max (32_768).
+     * Gives deep-review and patch-generation tools genuine headroom for
+     * complex multi-file diffs without switching to cost-unpredictable dynamic.
+     */
+    pro: 24_576,
 };
 const OUTPUT_TOKEN_BUDGET = {
-    flashTriage: 2_048,
-    flashTestPlan: 4_096,
+    flashTriage: 4_096,
+    /**
+     * Raised from 4_096: 15 test cases × pseudoCode@2_000 chars ≈ 7_500 tokens;
+     * staying at 4_096 risked MAX_TOKENS truncation on moderate test plans.
+     */
+    flashTestPlan: 8_192,
     flashApiBreaking: 4_096,
     flashComplexity: 2_048,
-    proReview: 8_192,
-    proPatch: 4_096,
+    /**
+     * Raised from 8_192: 25 findings × (title+explanation+recommendation) can
+     * exceed 8_192 tokens for rich, high-finding-count reviews.
+     */
+    proReview: 12_288,
+    /**
+     * Raised from 4_096: 10 search/replace blocks with multi-line code context
+     * can exceed the previous cap and cause MAX_TOKENS truncation.
+     */
+    proPatch: 8_192,
+};
+/**
+ * Per-task temperature presets for structured JSON generation.
+ * These are intentionally low: the model is already heavily constrained by
+ * the responseSchema, so lower temperatures improve schema-validation
+ * pass-through rates and reduce hallucinated field values.
+ */
+const TOOL_TEMPERATURE = {
+    /** Triage/classification tasks — deterministic structured extraction. */
+    triage: 0.1,
+    /** Analytical reasoning — consistent algorithmic analysis. */
+    analysis: 0.1,
+    /** Code patch generation — maximum precision for exact-match search blocks. */
+    patch: 0.0,
+    /** Test plan generation — allow modest diversity in test-case synthesis. */
+    creative: 0.2,
 };
 const DEFAULT_DETECT_HINT = 'detect';
-/** Thinking budget (tokens) for Flash model thinking tasks (test plans, search/replace). */
+/**
+ * Thinking budget (tokens) for Flash triage tools (impact, summary, API-breaking).
+ * Explicitly disabled (0) — these are classification/extraction tasks that do not
+ * benefit from a reasoning chain. Avoids default dynamic-thinking overhead.
+ * Flash 2.5 range: 0–24_576.
+ */
+export const FLASH_TRIAGE_THINKING_BUDGET = THINKING_BUDGET_TOKENS.flashTriage;
+/** Thinking budget (tokens) for Flash analysis tasks (test plans, complexity). */
 export const FLASH_THINKING_BUDGET = THINKING_BUDGET_TOKENS.flash;
-/** Thinking budget (tokens) for Pro model deep-analysis tasks (code quality inspection). */
+/** Thinking budget (tokens) for Pro model deep-analysis tasks (quality, patches). */
 export const PRO_THINKING_BUDGET = THINKING_BUDGET_TOKENS.pro;
 /** Output cap for Flash triage tools (impact, summary). */
 export const FLASH_TRIAGE_MAX_OUTPUT_TOKENS = OUTPUT_TOKEN_BUDGET.flashTriage;
@@ -33,6 +83,17 @@ export const PRO_PATCH_MAX_OUTPUT_TOKENS = OUTPUT_TOKEN_BUDGET.proPatch;
 export const FLASH_COMPLEXITY_MAX_OUTPUT_TOKENS = OUTPUT_TOKEN_BUDGET.flashComplexity;
 /** Extended timeout for Pro model calls (ms). Pro thinks longer than Flash. */
 export const DEFAULT_TIMEOUT_PRO_MS = 120_000;
+// ---------------------------------------------------------------------------
+// Temperature presets — see TOOL_TEMPERATURE constant for rationale.
+// ---------------------------------------------------------------------------
+/** Temperature for triage/classification tools (deterministic structured extraction). */
+export const TRIAGE_TEMPERATURE = TOOL_TEMPERATURE.triage;
+/** Temperature for analytical tools (consistent algorithmic reasoning). */
+export const ANALYSIS_TEMPERATURE = TOOL_TEMPERATURE.analysis;
+/** Temperature for code patch generation (maximum precision for search blocks). */
+export const PATCH_TEMPERATURE = TOOL_TEMPERATURE.patch;
+/** Temperature for creative synthesis tools (test plan generation). */
+export const CREATIVE_TEMPERATURE = TOOL_TEMPERATURE.creative;
 export const MODEL_TIMEOUT_MS = {
     defaultPro: DEFAULT_TIMEOUT_PRO_MS,
 };

package/dist/lib/tool-contracts.d.ts

@@ -1,3 +1,4 @@
+export declare const INSPECTION_FOCUS_AREAS: readonly ["security", "correctness", "performance", "regressions", "tests", "maintainability", "concurrency"];
 export interface ToolParameterContract {
     name: string;
     type: string;
@@ -15,6 +16,13 @@ export interface ToolContract {
     thinkingBudget?: number;
     /** Set to 0 for synchronous (non-Gemini) tools. */
     maxOutputTokens: number;
+    /**
+     * Sampling temperature for the Gemini call.
+     * Lower values (0.0–0.1) favour deterministic structured output;
+     * higher values (0.2) add diversity for creative synthesis tasks.
+     * Omit to use the global default (0.2).
+     */
+    temperature?: number;
     params: readonly ToolParameterContract[];
     outputShape: string;
     gotchas: readonly string[];
@@ -42,7 +50,9 @@ export declare const TOOL_CONTRACTS: readonly [{
     readonly purpose: "Assess severity, categories, breaking changes, and rollback complexity.";
     readonly model: "gemini-2.5-flash";
     readonly timeoutMs: 90000;
-    readonly maxOutputTokens: 2048;
+    readonly thinkingBudget: 0;
+    readonly maxOutputTokens: 4096;
+    readonly temperature: 0.1;
     readonly params: readonly [{
         readonly name: "repository";
         readonly type: "string";
@@ -64,7 +74,9 @@ export declare const TOOL_CONTRACTS: readonly [{
     readonly purpose: "Produce PR summary, risk rating, and merge recommendation.";
     readonly model: "gemini-2.5-flash";
     readonly timeoutMs: 90000;
-    readonly maxOutputTokens: 2048;
+    readonly thinkingBudget: 0;
+    readonly maxOutputTokens: 4096;
+    readonly temperature: 0.1;
     readonly params: readonly [{
         readonly name: "repository";
         readonly type: "string";
@@ -86,8 +98,9 @@ export declare const TOOL_CONTRACTS: readonly [{
     readonly purpose: "Deep code review with optional full-file context.";
     readonly model: "gemini-2.5-pro";
     readonly timeoutMs: 120000;
-    readonly thinkingBudget: 16384;
-    readonly maxOutputTokens: 8192;
+    readonly thinkingBudget: 24576;
+    readonly maxOutputTokens: 12288;
+    readonly temperature: 0.1;
     readonly params: readonly [{
         readonly name: "repository";
         readonly type: "string";
@@ -105,7 +118,7 @@ export declare const TOOL_CONTRACTS: readonly [{
         readonly type: "string[]";
         readonly required: false;
         readonly constraints: "1-12 items, 2-80 chars each";
-        readonly description: "Focused inspection categories.";
+        readonly description: `Focused inspection categories (e.g. ${string}).`;
     }, {
         readonly name: "maxFindings";
         readonly type: "number";
@@ -128,8 +141,9 @@ export declare const TOOL_CONTRACTS: readonly [{
     readonly purpose: "Generate verbatim search/replace fix blocks for one finding.";
     readonly model: "gemini-2.5-pro";
     readonly timeoutMs: 120000;
-    readonly thinkingBudget: 16384;
-    readonly maxOutputTokens: 4096;
+    readonly thinkingBudget: 24576;
+    readonly maxOutputTokens: 8192;
+    readonly temperature: 0;
     readonly params: readonly [{
         readonly name: "findingTitle";
         readonly type: "string";
@@ -152,8 +166,9 @@ export declare const TOOL_CONTRACTS: readonly [{
     readonly purpose: "Generate prioritized test cases and coverage guidance.";
     readonly model: "gemini-2.5-flash";
     readonly timeoutMs: 90000;
-    readonly thinkingBudget: 8192;
-    readonly maxOutputTokens: 4096;
+    readonly thinkingBudget: 16384;
+    readonly maxOutputTokens: 8192;
+    readonly temperature: 0.2;
     readonly params: readonly [{
         readonly name: "repository";
         readonly type: "string";
@@ -187,8 +202,9 @@ export declare const TOOL_CONTRACTS: readonly [{
     readonly purpose: "Analyze Big-O complexity and detect degradations in changed code.";
     readonly model: "gemini-2.5-flash";
     readonly timeoutMs: 90000;
-    readonly thinkingBudget: 8192;
+    readonly thinkingBudget: 16384;
     readonly maxOutputTokens: 2048;
+    readonly temperature: 0.1;
     readonly params: readonly [{
         readonly name: "language";
         readonly type: "string";
@@ -204,7 +220,9 @@ export declare const TOOL_CONTRACTS: readonly [{
     readonly purpose: "Detect breaking API/interface changes in a diff.";
     readonly model: "gemini-2.5-flash";
     readonly timeoutMs: 90000;
+    readonly thinkingBudget: 0;
     readonly maxOutputTokens: 4096;
+    readonly temperature: 0.1;
     readonly params: readonly [{
         readonly name: "language";
         readonly type: "string";

package/dist/lib/tool-contracts.js

@@ -1,5 +1,14 @@
-import { DEFAULT_TIMEOUT_PRO_MS, FLASH_API_BREAKING_MAX_OUTPUT_TOKENS, FLASH_COMPLEXITY_MAX_OUTPUT_TOKENS, FLASH_MODEL, FLASH_TEST_PLAN_MAX_OUTPUT_TOKENS, FLASH_THINKING_BUDGET, FLASH_TRIAGE_MAX_OUTPUT_TOKENS, PRO_MODEL, PRO_PATCH_MAX_OUTPUT_TOKENS, PRO_REVIEW_MAX_OUTPUT_TOKENS, PRO_THINKING_BUDGET, } from './model-config.js';
+import { ANALYSIS_TEMPERATURE, CREATIVE_TEMPERATURE, DEFAULT_TIMEOUT_PRO_MS, FLASH_API_BREAKING_MAX_OUTPUT_TOKENS, FLASH_COMPLEXITY_MAX_OUTPUT_TOKENS, FLASH_MODEL, FLASH_TEST_PLAN_MAX_OUTPUT_TOKENS, FLASH_THINKING_BUDGET, FLASH_TRIAGE_MAX_OUTPUT_TOKENS, FLASH_TRIAGE_THINKING_BUDGET, PATCH_TEMPERATURE, PRO_MODEL, PRO_PATCH_MAX_OUTPUT_TOKENS, PRO_REVIEW_MAX_OUTPUT_TOKENS, PRO_THINKING_BUDGET, TRIAGE_TEMPERATURE, } from './model-config.js';
 const DEFAULT_TIMEOUT_FLASH_MS = 90_000;
+export const INSPECTION_FOCUS_AREAS = [
+    'security',
+    'correctness',
+    'performance',
+    'regressions',
+    'tests',
+    'maintainability',
+    'concurrency',
+];
 export const TOOL_CONTRACTS = [
     {
         name: 'generate_diff',
@@ -31,7 +40,9 @@ export const TOOL_CONTRACTS = [
         purpose: 'Assess severity, categories, breaking changes, and rollback complexity.',
         model: FLASH_MODEL,
         timeoutMs: DEFAULT_TIMEOUT_FLASH_MS,
+        thinkingBudget: FLASH_TRIAGE_THINKING_BUDGET,
         maxOutputTokens: FLASH_TRIAGE_MAX_OUTPUT_TOKENS,
+        temperature: TRIAGE_TEMPERATURE,
         params: [
             {
                 name: 'repository',
@@ -62,7 +73,9 @@ export const TOOL_CONTRACTS = [
         purpose: 'Produce PR summary, risk rating, and merge recommendation.',
         model: FLASH_MODEL,
         timeoutMs: DEFAULT_TIMEOUT_FLASH_MS,
+        thinkingBudget: FLASH_TRIAGE_THINKING_BUDGET,
         maxOutputTokens: FLASH_TRIAGE_MAX_OUTPUT_TOKENS,
+        temperature: TRIAGE_TEMPERATURE,
         params: [
             {
                 name: 'repository',
@@ -95,6 +108,7 @@ export const TOOL_CONTRACTS = [
         timeoutMs: DEFAULT_TIMEOUT_PRO_MS,
         thinkingBudget: PRO_THINKING_BUDGET,
         maxOutputTokens: PRO_REVIEW_MAX_OUTPUT_TOKENS,
+        temperature: ANALYSIS_TEMPERATURE,
         params: [
             {
                 name: 'repository',
@@ -115,7 +129,7 @@ export const TOOL_CONTRACTS = [
                 type: 'string[]',
                 required: false,
                 constraints: '1-12 items, 2-80 chars each',
-                description: 'Focused inspection categories.',
+                description: `Focused inspection categories (e.g. ${INSPECTION_FOCUS_AREAS.join(', ')}).`,
             },
             {
                 name: 'maxFindings',
@@ -151,6 +165,7 @@ export const TOOL_CONTRACTS = [
         timeoutMs: DEFAULT_TIMEOUT_PRO_MS,
         thinkingBudget: PRO_THINKING_BUDGET,
         maxOutputTokens: PRO_PATCH_MAX_OUTPUT_TOKENS,
+        temperature: PATCH_TEMPERATURE,
         params: [
             {
                 name: 'findingTitle',
@@ -185,6 +200,7 @@ export const TOOL_CONTRACTS = [
         timeoutMs: DEFAULT_TIMEOUT_FLASH_MS,
         thinkingBudget: FLASH_THINKING_BUDGET,
         maxOutputTokens: FLASH_TEST_PLAN_MAX_OUTPUT_TOKENS,
+        temperature: CREATIVE_TEMPERATURE,
         params: [
             {
                 name: 'repository',
@@ -231,6 +247,7 @@ export const TOOL_CONTRACTS = [
         timeoutMs: DEFAULT_TIMEOUT_FLASH_MS,
         thinkingBudget: FLASH_THINKING_BUDGET,
         maxOutputTokens: FLASH_COMPLEXITY_MAX_OUTPUT_TOKENS,
+        temperature: ANALYSIS_TEMPERATURE,
         params: [
             {
                 name: 'language',
@@ -252,7 +269,9 @@ export const TOOL_CONTRACTS = [
         purpose: 'Detect breaking API/interface changes in a diff.',
         model: FLASH_MODEL,
         timeoutMs: DEFAULT_TIMEOUT_FLASH_MS,
+        thinkingBudget: FLASH_TRIAGE_THINKING_BUDGET,
         maxOutputTokens: FLASH_API_BREAKING_MAX_OUTPUT_TOKENS,
+        temperature: TRIAGE_TEMPERATURE,
         params: [
             {
                 name: 'language',

package/dist/lib/tool-factory.d.ts

@@ -63,6 +63,13 @@ export interface StructuredToolTaskConfig<TInput extends object = Record<string,
     timeoutMs?: number;
     /** Optional max output tokens for Gemini. */
     maxOutputTokens?: number;
+    /**
+     * Optional sampling temperature for this tool's Gemini call.
+     * Lower values (0.0–0.1) favour determinism for structured extraction;
+     * higher values (0.2) add useful diversity for creative synthesis tasks.
+     * Falls back to the global default (0.2) when omitted.
+     */
+    temperature?: number;
     /** Optional opt-in to Gemini thought output. Defaults to false. */
     includeThoughts?: boolean;
     /** Optional formatter for human-readable text output. */

package/dist/lib/tool-factory.js

@@ -40,6 +40,9 @@ function createGenerationRequest(config, promptParts, responseSchema, onLog, sig
     if (config.maxOutputTokens !== undefined) {
         request.maxOutputTokens = config.maxOutputTokens;
     }
+    if (config.temperature !== undefined) {
+        request.temperature = config.temperature;
+    }
     if (config.includeThoughts !== undefined) {
         request.includeThoughts = config.includeThoughts;
     }

package/dist/prompts/index.d.ts

@@ -1,2 +1,11 @@
 import type { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
+export declare const PROMPT_DEFINITIONS: readonly [{
+    readonly name: "get-help";
+    readonly title: "Get Help";
+    readonly description: "Server instructions.";
+}, {
+    readonly name: "review-guide";
+    readonly title: "Review Guide";
+    readonly description: "Workflow guide for tool/focus area.";
+}];
 export declare function registerAllPrompts(server: McpServer, instructions: string): void;

package/dist/prompts/index.js

@@ -1,20 +1,19 @@
 import { completable } from '@modelcontextprotocol/sdk/server/completable.js';
 import { z } from 'zod';
-import { getToolContract, getToolContractNames, } from '../lib/tool-contracts.js';
-const HELP_PROMPT_NAME = 'get-help';
-const HELP_PROMPT_TITLE = 'Get Help';
-const HELP_PROMPT_DESCRIPTION = 'Server instructions.';
-const REVIEW_GUIDE_PROMPT_NAME = 'review-guide';
-const REVIEW_GUIDE_PROMPT_TITLE = 'Review Guide';
-const REVIEW_GUIDE_PROMPT_DESCRIPTION = 'Workflow guide for tool/focus area.';
-const TOOLS = getToolContractNames();
-const FOCUS_AREAS = [
-    'security',
-    'correctness',
-    'performance',
-    'regressions',
-    'tests',
+import { getToolContract, getToolContractNames, INSPECTION_FOCUS_AREAS, } from '../lib/tool-contracts.js';
+export const PROMPT_DEFINITIONS = [
+    {
+        name: 'get-help',
+        title: 'Get Help',
+        description: 'Server instructions.',
+    },
+    {
+        name: 'review-guide',
+        title: 'Review Guide',
+        description: 'Workflow guide for tool/focus area.',
+    },
 ];
+const TOOLS = getToolContractNames();
 const TOOL_DESCRIPTION_TEXT = 'Select tool for review guide.';
 const FOCUS_DESCRIPTION_TEXT = 'Select focus area.';
 const FOCUS_AREA_GUIDES = {
@@ -23,6 +22,8 @@ const FOCUS_AREA_GUIDES = {
     performance: 'Focus: Complexity, allocations, I/O, queries.',
     regressions: 'Focus: Behavior changes, guards, types, breaks.',
     tests: 'Focus: Coverage, edge cases, flakes, error paths.',
+    maintainability: 'Focus: Complexity, readability, structure, patterns.',
+    concurrency: 'Focus: Race conditions, deadlocks, shared state.',
 };
 function completeByPrefix(values, prefix) {
     const matches = [];
@@ -52,11 +53,12 @@ function getFocusAreaGuide(focusArea) {
     return getGuide(FOCUS_AREA_GUIDES, focusArea, (area) => `Focus on ${area} concerns.`);
 }
 function registerHelpPrompt(server, instructions) {
-    server.registerPrompt(HELP_PROMPT_NAME, {
-        title: HELP_PROMPT_TITLE,
-        description: HELP_PROMPT_DESCRIPTION,
+    const def = PROMPT_DEFINITIONS[0];
+    server.registerPrompt(def.name, {
+        title: def.title,
+        description: def.description,
     }, () => ({
-        description: HELP_PROMPT_DESCRIPTION,
+        description: def.description,
         messages: [
             {
                 role: 'user',
@@ -85,12 +87,13 @@ function buildReviewGuideText(tool, focusArea) {
         `> Tip: Run \`get-help\` for full server documentation.`);
 }
 function registerReviewGuidePrompt(server) {
-    server.registerPrompt(REVIEW_GUIDE_PROMPT_NAME, {
-        title: REVIEW_GUIDE_PROMPT_TITLE,
-        description: REVIEW_GUIDE_PROMPT_DESCRIPTION,
+    const def = PROMPT_DEFINITIONS[1];
+    server.registerPrompt(def.name, {
+        title: def.title,
+        description: def.description,
         argsSchema: {
             tool: completable(z.string().describe(TOOL_DESCRIPTION_TEXT), (value) => completeByPrefix(TOOLS, value)),
-            focusArea: completable(z.string().describe(FOCUS_DESCRIPTION_TEXT), (value) => completeByPrefix(FOCUS_AREAS, value)),
+            focusArea: completable(z.string().describe(FOCUS_DESCRIPTION_TEXT), (value) => completeByPrefix(INSPECTION_FOCUS_AREAS, value)),
         },
     }, ({ tool, focusArea }) => ({
         description: `Code review guide: ${tool} / ${focusArea}`,

package/dist/resources/index.d.ts

@@ -1,2 +1,12 @@
 import { type McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
+export interface StaticResourceDef {
+    id: string;
+    uri: string;
+    title: string;
+    description: string;
+    priority: number;
+    content: () => string;
+}
+export declare const STATIC_RESOURCES: readonly StaticResourceDef[];
+export declare const DIFF_RESOURCE_DESCRIPTION = "The most recently generated diff, cached by generate_diff. Read by all review tools automatically.";
 export declare function registerAllResources(server: McpServer, instructions: string): void;

package/dist/resources/index.js

@@ -9,7 +9,7 @@ const RESOURCE_AUDIENCE = ['assistant'];
 function createMarkdownContent(uri, text) {
     return { uri: uri.href, mimeType: RESOURCE_MIME_TYPE, text };
 }
-const STATIC_RESOURCES = [
+export const STATIC_RESOURCES = [
     {
         id: 'server-instructions',
         uri: 'internal://instructions',
@@ -77,10 +77,11 @@ function registerToolInfoResources(server) {
         return { contents: [createMarkdownContent(uri, text)] };
     });
 }
+export const DIFF_RESOURCE_DESCRIPTION = 'The most recently generated diff, cached by generate_diff. Read by all review tools automatically.';
 function registerDiffResource(server) {
     server.registerResource('diff-current', new ResourceTemplate(DIFF_RESOURCE_URI, { list: undefined }), {
         title: 'Current Diff',
-        description: 'The most recently generated diff, cached by generate_diff. Read by all review tools automatically.',
+        description: DIFF_RESOURCE_DESCRIPTION,
         mimeType: 'text/x-patch',
         annotations: {
             audience: ['assistant'],

package/dist/resources/instructions.js

@@ -1,16 +1,12 @@
 import { getToolContracts } from '../lib/tool-contracts.js';
+import { PROMPT_DEFINITIONS } from '../prompts/index.js';
+import { DIFF_RESOURCE_DESCRIPTION, STATIC_RESOURCES } from './index.js';
 import { getSharedConstraints } from './tool-info.js';
-const PROMPT_LIST = [
-    '- `get-help`: Returns these server instructions.',
-    '- `review-guide`: Workflow guide for a selected tool and focus area.',
-];
+const PROMPT_LIST = PROMPT_DEFINITIONS.map((def) => `- \`${def.name}\`: ${def.description}`);
 const RESOURCE_LIST = [
-    '- `internal://instructions`: This document.',
-    '- `internal://tool-catalog`: Tool matrix and cross-tool data flow.',
-    '- `internal://workflows`: Recommended multi-step tool workflows.',
-    '- `internal://server-config`: Runtime limits and model configuration.',
+    ...STATIC_RESOURCES.map((def) => `- \`${def.uri}\`: ${def.description}`),
     '- `internal://tool-info/{toolName}`: Per-tool contract details.',
-    '- `diff://current`: Cached diff from the most recent generate_diff run.',
+    `- \`diff://current\`: ${DIFF_RESOURCE_DESCRIPTION}`,
 ];
 function formatParameterLine(parameter) {
     const required = parameter.required ? 'required' : 'optional';

package/dist/schemas/inputs.js

@@ -51,7 +51,7 @@ export const InspectCodeQualityInputSchema = z.strictObject({
         .min(1)
         .max(INPUT_LIMITS.focusArea.maxItems)
         .optional()
-        .describe('Review focus areas. Standard tags: security, performance, correctness, maintainability, concurrency. Omit for general review.'),
+        .describe('Review focus areas. Standard tags: security, correctness, performance, regressions, tests, maintainability, concurrency. Omit for general review.'),
     maxFindings: createOptionalBoundedInteger(INPUT_LIMITS.maxFindings.min, INPUT_LIMITS.maxFindings.max, 'Max findings (1-25). Default: 10.'),
     files: z
         .array(FileContextSchema)

package/dist/tools/analyze-complexity.js

@@ -26,6 +26,9 @@ export function registerAnalyzeComplexityTool(server) {
         ...(TOOL_CONTRACT.thinkingBudget !== undefined
             ? { thinkingBudget: TOOL_CONTRACT.thinkingBudget }
             : undefined),
+        ...(TOOL_CONTRACT.temperature !== undefined
+            ? { temperature: TOOL_CONTRACT.temperature }
+            : undefined),
         validateInput: (_input, ctx) => {
             const slot = ctx.diffSlot;
             if (!slot)

package/dist/tools/analyze-pr-impact.js

@@ -27,6 +27,12 @@ export function registerAnalyzePrImpactTool(server) {
         model: TOOL_CONTRACT.model,
         timeoutMs: TOOL_CONTRACT.timeoutMs,
         maxOutputTokens: TOOL_CONTRACT.maxOutputTokens,
+        ...(TOOL_CONTRACT.thinkingBudget !== undefined
+            ? { thinkingBudget: TOOL_CONTRACT.thinkingBudget }
+            : undefined),
+        ...(TOOL_CONTRACT.temperature !== undefined
+            ? { temperature: TOOL_CONTRACT.temperature }
+            : undefined),
         validateInput: (_input, ctx) => {
             const slot = ctx.diffSlot;
             if (!slot)

package/dist/tools/detect-api-breaking.js

@@ -23,6 +23,12 @@ export function registerDetectApiBreakingTool(server) {
         model: TOOL_CONTRACT.model,
         timeoutMs: TOOL_CONTRACT.timeoutMs,
         maxOutputTokens: TOOL_CONTRACT.maxOutputTokens,
+        ...(TOOL_CONTRACT.thinkingBudget !== undefined
+            ? { thinkingBudget: TOOL_CONTRACT.thinkingBudget }
+            : undefined),
+        ...(TOOL_CONTRACT.temperature !== undefined
+            ? { temperature: TOOL_CONTRACT.temperature }
+            : undefined),
         validateInput: (_input, ctx) => {
             const slot = ctx.diffSlot;
             if (!slot)

package/dist/tools/generate-review-summary.js

@@ -34,6 +34,12 @@ export function registerGenerateReviewSummaryTool(server) {
         model: TOOL_CONTRACT.model,
         timeoutMs: TOOL_CONTRACT.timeoutMs,
         maxOutputTokens: TOOL_CONTRACT.maxOutputTokens,
+        ...(TOOL_CONTRACT.thinkingBudget !== undefined
+            ? { thinkingBudget: TOOL_CONTRACT.thinkingBudget }
+            : undefined),
+        ...(TOOL_CONTRACT.temperature !== undefined
+            ? { temperature: TOOL_CONTRACT.temperature }
+            : undefined),
         validateInput: (_input, ctx) => {
             const slot = ctx.diffSlot;
             if (!slot)

package/dist/tools/generate-test-plan.js

@@ -30,6 +30,9 @@ export function registerGenerateTestPlanTool(server) {
         ...(TOOL_CONTRACT.thinkingBudget !== undefined
             ? { thinkingBudget: TOOL_CONTRACT.thinkingBudget }
             : undefined),
+        ...(TOOL_CONTRACT.temperature !== undefined
+            ? { temperature: TOOL_CONTRACT.temperature }
+            : undefined),
         validateInput: (_input, ctx) => {
             const slot = ctx.diffSlot;
             if (!slot)

package/dist/tools/inspect-code-quality.js

@@ -66,6 +66,9 @@ export function registerInspectCodeQualityTool(server) {
         ...(TOOL_CONTRACT.thinkingBudget !== undefined
             ? { thinkingBudget: TOOL_CONTRACT.thinkingBudget }
             : undefined),
+        ...(TOOL_CONTRACT.temperature !== undefined
+            ? { temperature: TOOL_CONTRACT.temperature }
+            : undefined),
         progressContext: (input) => {
             const fileCount = input.files?.length;
             return fileCount ? `+${fileCount} files` : '';

package/dist/tools/suggest-search-replace.js

@@ -30,6 +30,9 @@ export function registerSuggestSearchReplaceTool(server) {
         ...(TOOL_CONTRACT.thinkingBudget !== undefined
             ? { thinkingBudget: TOOL_CONTRACT.thinkingBudget }
             : undefined),
+        ...(TOOL_CONTRACT.temperature !== undefined
+            ? { temperature: TOOL_CONTRACT.temperature }
+            : undefined),
         validateInput: (_input, ctx) => {
             const slot = ctx.diffSlot;
             if (!slot)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@j0hanz/code-review-analyst-mcp",
-  "version": "1.4.2",
+  "version": "1.4.4",
   "mcpName": "io.github.j0hanz/code-review-analyst",
   "description": "Gemini-powered MCP server for code review analysis.",
   "type": "module",