@softerist/heuristic-mcp 3.0.3 → 3.0.6

package/.agent/workflows/code-review.md

@@ -7,6 +7,7 @@ last_updated: 2026-01-29
 You are a senior software engineer and code reviewer. Review the code provided according to the following strict guidelines.
 
 ## 1. Review Scope & Context
+
 - **For code <100 lines**: Quick review focusing on correctness and obvious bugs.
 - **For code 100-500 lines**: Standard review following all sections below.
 - **For code >500 lines or critical systems**: Deep review with extra emphasis on architecture, security, and edge cases.
@@ -14,46 +15,56 @@ You are a senior software engineer and code reviewer. Review the code provided a
 - **Domain-Specific Checks**: If the code domain is identifiable (e.g., web, mobile, embedded, ML, healthcare), apply relevant domain-specific standards (e.g., WCAG for web, PII handling for healthcare, real-time constraints for embedded).
 
 ## 2. Line-by-Line Review
+
 - Review every function and critical code path. For files >200 lines, focus on functions with issues rather than every line.
 - Group line-by-line findings by severity (**Critical** first), then by file/function within each severity tier.
 - For each issue, quote the relevant line(s).
 - **Confidence Ratings**: For ambiguous issues where context is missing, explicitly note `[Confidence: Low/Medium/High]` and explain what additional information would clarify the finding.
 
 ## 3. High-Level Review
+
 - Focus on system-level patterns not visible at function scope: module organization, layer violations, missing abstraction layers, deployment concerns.
 - Do not check for issues already covered in line-by-line (avoid redundancy).
 
 ## 4. Follow-up Reviews
+
 - For follow-up reviews, focus on: regression verification, new code introduced by fixes, and cross-cutting impact of changes.
 
 ## 5. Stalled/Incomplete Logic
+
 - Check for "forgotten error paths": places where exceptions/errors could occur but aren't caught, or catch blocks that are empty/only log without recovery strategy.
 - Verify TODOs or placeholders.
 
 ## 6. Consistency
+
 - Identify the dominant pattern in the codebase and flag deviations. If no clear majority, note the inconsistency itself as an issue.
 - Check naming conventions, coding style, and file structure.
 
 ## 7. Fix Plan
+
 Provide a prioritized checklist (most critical first):
+
 - **Estimated effort per item** using this scale:
-  * **S (Small)**: <2 hours, isolated change, no breaking changes.
-  * **M (Medium)**: 2-8 hours, may affect multiple files, local refactoring.
-  * **L (Large)**: >8 hours, architectural changes, breaking changes, or requires team coordination.
+  - **S (Small)**: <2 hours, isolated change, no breaking changes.
+  - **M (Medium)**: 2-8 hours, may affect multiple files, local refactoring.
+  - **L (Large)**: >8 hours, architectural changes, breaking changes, or requires team coordination.
 - For each item, note any dependencies: "Requires #N" if another fix must be completed first.
 - Group independent fixes together to enable parallel work.
 
 ## 8. Patch
+
 - Propose concrete code edits in **unified diff format** for **ALL critical severity issues**.
 - For **high severity** issues, provide patches for the top 5 most impactful.
 - For new files or large refactors (>50 lines changed), provide complete code blocks with clear before/after markers.
 
 ## 9. Tests
+
 - Separate **unit tests** (isolated, mocked dependencies) from **integration tests** (real dependencies). Specify which category each test belongs to.
 - List the exact tests you'd add (test names + what each asserts).
 - For performance issues, include benchmark or load tests with acceptable thresholds.
 
 ## Output Format Constraints
+
 - **Do not skip sections.** If a section has no findings, state "No issues found" and briefly explain why (e.g., "No concurrency issues: code is single-threaded").
 - **Assume production context**: Treat as a professional audit. If context is unclear, state assumptions (e.g., "assuming user-facing web application with moderate traffic").
 - **Be exhaustive and specific.** Do not be polite; be direct.

package/LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2025 Omar Haris
+Copyright (c) 2026 Omar Haris, softerist
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

package/config.jsonc

@@ -563,7 +563,7 @@
     "gitattributes",
     "editorconfig",
     "npmrc",
-    "yarnrc"
+    "yarnrc",
   ],
   // Specific filenames to include regardless of extension.
   "fileNames": [
@@ -612,7 +612,7 @@
     "docker-compose.yml",
     "docker-compose.yaml",
     "makefile.am",
-    "makefile.in"
+    "makefile.in",
   ],
   // Glob patterns to exclude from indexing.
   "excludePatterns": [
@@ -624,7 +624,7 @@
     "**/.next/**",
     "**/target/**",
     "**/vendor/**",
-    "**/.smart-coding-cache/**"
+    "**/.smart-coding-cache/**",
   ],
   // Enable project-type detection + smart ignore patterns.
   "smartIndexing": true,
@@ -719,5 +719,5 @@
   // Persist ANN index to disk.
   "annIndexCache": true,
   // ANN distance metric.
-  "annMetric": "cosine"
+  "annMetric": "cosine",
 }

package/debug-pids.js

@@ -1,27 +1,30 @@
-import { exec } from 'child_process';
-import util from 'util';
-
-const execPromise = util.promisify(exec);
-
-async function check() {
+import { exec } from 'child_process';
+import util from 'util';
+
+const execPromise = util.promisify(exec);
+
+async function check() {
   console.info('Checking processes...');
-  try {
-    const { stdout } = await execPromise(
-      `powershell -NoProfile -Command "Get-CimInstance Win32_Process | Where-Object { $_.Name -eq 'node.exe' } | Select-Object ProcessId, CommandLine, ParentProcessId | ConvertTo-Json"`
-    );
-    const processes = JSON.parse(stdout);
-    const list = Array.isArray(processes) ? processes : [processes];
-    
-    for (const p of list) {
-      if (p.CommandLine && (p.CommandLine.includes('heuristic-mcp') || p.CommandLine.includes('index.js'))) {
+  try {
+    const { stdout } = await execPromise(
+      `powershell -NoProfile -Command "Get-CimInstance Win32_Process | Where-Object { $_.Name -eq 'node.exe' } | Select-Object ProcessId, CommandLine, ParentProcessId | ConvertTo-Json"`
+    );
+    const processes = JSON.parse(stdout);
+    const list = Array.isArray(processes) ? processes : [processes];
+
+    for (const p of list) {
+      if (
+        p.CommandLine &&
+        (p.CommandLine.includes('heuristic-mcp') || p.CommandLine.includes('index.js'))
+      ) {
         console.info(`PID: ${p.ProcessId}, Parent: ${p.ParentProcessId}`);
         console.info(`CMD: ${p.CommandLine}`);
         console.info('---');
-      }
-    }
-  } catch (err) {
-    console.error('Error:', err.message);
-  }
-}
-
-check();
+      }
+    }
+  } catch (err) {
+    console.error('Error:', err.message);
+  }
+}
+
+check();

package/features/find-similar-code.js

@@ -48,7 +48,7 @@ export class FindSimilarCode {
     // Check if input is too large and truncate intelligently
     const estimatedTokens = estimateTokens(code);
     const limit = getModelTokenLimit(this.config.embeddingModel);
-    
+
     // If input is significantly larger than the model limit, we should chunk it
     if (estimatedTokens > limit) {
       // Use smartChunk to get a semantically valid first block
@@ -66,14 +66,16 @@ export class FindSimilarCode {
       pooling: 'mean',
       normalize: true,
     });
-    const codeVector = codeEmbed.data; // Keep as Float32Array for performance
-    const codeVectorTyped = codeVector;
+
+    // CRITICAL: Deep copy Float32Array to avoid detachment issues with WASM/Workers
+    // accessing a detached buffer from a reusable ONNX tensor can crash the process.
+    const codeVector = new Float32Array(codeEmbed.data);
 
     let candidates = vectorStore;
     let usedAnn = false;
     if (this.config.annEnabled) {
       const candidateCount = this.getAnnCandidateCount(maxResults, vectorStore.length);
-      const annLabels = await this.cache.queryAnn(codeVectorTyped, candidateCount);
+      const annLabels = await this.cache.queryAnn(codeVector, candidateCount);
       if (annLabels && annLabels.length >= maxResults) {
         usedAnn = true;
         const seen = new Set();
@@ -88,27 +90,27 @@ export class FindSimilarCode {
     }
 
     const normalizedInput = codeToEmbed.trim().replace(/\s+/g, ' ');
-    
+
     /**
      * Batch scoring function to prevent blocking the event loop
      */
     const scoreAndFilter = async (chunks) => {
       const BATCH_SIZE = 500;
       const scored = [];
-      
+
       for (let i = 0; i < chunks.length; i += BATCH_SIZE) {
         const batch = chunks.slice(i, i + BATCH_SIZE);
-        
+
         // Yield to event loop between batches
         if (i > 0) {
           await new Promise((resolve) => setTimeout(resolve, 0));
         }
-        
+
         for (const chunk of batch) {
           const vector = this.getChunkVector(chunk);
           if (!vector) continue;
           const similarity = dotSimilarity(codeVector, vector);
-          
+
           if (similarity >= minSimilarity) {
             // Deduplicate against input
             if (normalizedInput) {
@@ -116,17 +118,17 @@ export class FindSimilarCode {
               const normalizedChunk = content.trim().replace(/\s+/g, ' ');
               if (normalizedChunk === normalizedInput) continue;
             }
-            
+
             scored.push({ ...chunk, similarity });
           }
         }
       }
-      
+
       return scored.sort((a, b) => b.similarity - a.similarity);
     };
 
     let filteredResults = await scoreAndFilter(candidates);
-    
+
     // Fallback to full scan if ANN didn't provide enough results
     // Optimization: Skip full scan on large codebases to avoid long pauses
     const MAX_FULL_SCAN_SIZE = 5000;
@@ -143,12 +145,14 @@ export class FindSimilarCode {
           return { ...chunk, content: await this.getChunkContent(chunk) };
         }
         return chunk;
-      }),
+      })
     );
 
     return {
       results,
-      message: warningMessage || (results.length === 0 ? 'No similar code found above the similarity threshold.' : null),
+      message:
+        warningMessage ||
+        (results.length === 0 ? 'No similar code found above the similarity threshold.' : null),
     };
   }
 
@@ -160,7 +164,7 @@ export class FindSimilarCode {
     const formatted = await Promise.all(
       results.map(async (r, idx) => {
         const relPath = path.relative(this.config.searchDirectory, r.file);
-        const content = r.content ?? await this.getChunkContent(r);
+        const content = r.content ?? (await this.getChunkContent(r));
         return (
           `## Similar Code ${idx + 1} (Similarity: ${(r.similarity * 100).toFixed(1)}%)\n` +
           `**File:** \`${relPath}\`\n` +
@@ -172,7 +176,7 @@ export class FindSimilarCode {
           '\n' +
           '```\n'
         );
-      }),
+      })
     );
 
     return formatted.join('\n');