chekk 0.4.0 → 0.4.1

package/bin/chekk.js

@@ -4,7 +4,7 @@ import { execSync, spawn } from 'child_process';
 import { Command } from 'commander';
 import { run } from '../src/index.js';
 
-const LOCAL_VERSION = '0.4.0';
+const LOCAL_VERSION = '0.4.1';
 
 // ── Auto-update check ──
 // If running from a cached npx install, check if there's a newer version

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "chekk",
-  "version": "0.4.0",
+  "version": "0.4.1",
   "description": "See how you prompt. Chekk analyzes your AI coding workflow and tells you what kind of engineer you are.",
   "bin": {
     "chekk": "./bin/chekk.js"

package/src/display.js

@@ -80,10 +80,32 @@ function displayLabeledSnippet(label, prompt, maxLen = 120) {
   console.log(`  ${dim('\u21B3')} ${dim(label + ':')} ${dim.italic('\u201C' + s + '\u201D')}`);
 }
 
+// Cross-dimension filters: reject prompts that clearly belong to another dimension
+const architecturalRe = /\b(architect|design|refactor|redesign|restructure|system design|data model|schema|api design|infrastructure|migration|strategy)\b/i;
+const debugRe = /\b(error|bug|broken|crash|fail|exception|traceback|stack trace|doesn'?t work|not working|TypeError|SyntaxError|ImportError|ReferenceError|500|502|503|404|CORS)\b/i;
+const planningRe = /\b(plan|breakdown|break down|think through|help me think|pros and cons|how should|code review|audit)\b/i;
+
+// For each dimension, prompts matching these patterns are *excluded* as evidence
+const dimensionExclusions = {
+  'specific_report': [architecturalRe, planningRe],
+  'quick_fix': [architecturalRe, planningRe],
+  'architectural': [debugRe],
+  'planning': [debugRe],
+  'exploratory': [debugRe],
+  'decomposition': [],
+  'followup': [],
+  'context_setting': [],
+  'refinement': [],
+};
+
 function pickExample(examples, type) {
   if (!examples || !examples.length) return null;
-  const match = examples.find(e => e.type === type);
-  return match ? match.prompt : null;
+  const exclusions = dimensionExclusions[type] || [];
+  // Prefer a match that doesn't trigger exclusion patterns
+  const candidates = examples.filter(e => e.type === type);
+  if (candidates.length === 0) return null;
+  const clean = candidates.find(e => !exclusions.some(re => re.test(e.prompt)));
+  return (clean || candidates[0]).prompt;
 }
 
 // ── Box drawing ──
@@ -128,7 +150,7 @@ export function displayHeader() {
   console.log();
   const lines = [
     '',
-    `   ${bold.white('chekk')}${dim(' v0.4.0')}`,
+    `   ${bold.white('chekk')}${dim(' v0.4.1')}`,
     `   ${dim('engineering capability profile')}`,
     '',
   ];
@@ -186,7 +208,7 @@ function displayProfileHeader(result, extra = {}) {
   console.log(`  ${bold.white('ENGINEERING CAPABILITY PROFILE')}`);
   console.log();
   if (sessionStats) {
-    console.log(`  ${dim(`Generated ${dateStr} | chekk v0.4.0`)}`);
+    console.log(`  ${dim(`Generated ${dateStr} | chekk v0.4.1`)}`);
     console.log(`  ${dim(`Analysis: ${sessionStats.totalSessions} sessions \u00B7 ${sessionStats.tools.length} tool${sessionStats.tools.length > 1 ? 's' : ''} \u00B7 ${numberFormat(sessionStats.totalExchanges)} exchanges`)}`);
     if (sessionStats.dateRangeShort) {
       console.log(`  ${dim(`Period: ${sessionStats.dateRangeShort}`)}`);

package/src/insights.js

@@ -417,11 +417,11 @@ export function generateAssessment(result, metrics, signatures, watchPoints) {
   const weakest = dims[dims.length - 1];
 
   // Build assessment parts
-  let assessment = `This engineer demonstrates ${dimQualitative(strongest.score)} ${strongest.label}`;
+  let assessment = `This engineer demonstrates ${dimQualitative(strongest.score).toLowerCase()} ${strongest.label}`;
 
   // Add signature mention if available
   if (signatures.length > 0) {
-    assessment += ` with a distinctive pattern of ${signatures[0].name.toLowerCase()}`;
+    assessment += ` with a distinctive pattern of ${formatSignatureName(signatures[0].name)}`;
   }
   assessment += '.';
 
@@ -466,6 +466,14 @@ function dimQualitative(score) {
   return 'Early-stage';
 }
 
+// Lowercase a signature name for prose while preserving acronyms like "AI", "TDD"
+function formatSignatureName(name) {
+  return name
+    .toLowerCase()
+    .replace(/\bai\b/g, 'AI')
+    .replace(/\btdd\b/g, 'TDD');
+}
+
 // ══════════════════════════════════════════════
 // CONFIDENCE — Data volume indicator
 // ══════════════════════════════════════════════

package/src/metrics/ai-leverage.js

@@ -11,6 +11,16 @@
  * - Diversity of tool usage (not just "write code" but also explore, analyze, test)
  */
 
+// ── Evidence quality filter ──
+const noisePatterns = /^This session is being continued|^\[?[0-9T:.Z-]{20,}|^\S+@\S+.*[%$#>]|^\s*\$\s|^\s*>\s/;
+function isGoodEvidence(prompt) {
+  if (!prompt || prompt.length < 40 || prompt.length > 600) return false;
+  if (noisePatterns.test(prompt)) return false;
+  const alpha = prompt.replace(/[^a-zA-Z]/g, '').length;
+  if (alpha / prompt.length < 0.4) return false;
+  return true;
+}
+
 const architecturalPatterns = /\b(architect|design|refactor|redesign|restructure|system design|data model|schema|api design|interface|abstract|pattern|trade-?off|scalab|approach|strategy|migration|infrastructure)\b/i;
 const planningPatterns = /\b(plan|breakdown|break down|think through|help me think|what('?s| is) the best (way|approach)|how should (i|we)|pros and cons|options for|compare|evaluate|review my|code review|audit)\b/i;
 const exploratoryPatterns = /\b(explain|understand|how does|what does|why does|walk me through|investigate|diagnose|analyze|explore|deep dive|look into)\b/i;
@@ -53,15 +63,15 @@ export function computeAILeverage(sessions) {
       // Categorize prompt type
       if (architecturalPatterns.test(prompt)) {
         architecturalPrompts++;
-        if (prompt.length > bestArchLen) { bestArchLen = prompt.length; bestArchPrompt = prompt; }
+        if (isGoodEvidence(prompt) && prompt.length > bestArchLen) { bestArchLen = prompt.length; bestArchPrompt = prompt; }
       }
       if (planningPatterns.test(prompt)) {
         planningPrompts++;
-        if (prompt.length > bestPlanLen) { bestPlanLen = prompt.length; bestPlanPrompt = prompt; }
+        if (isGoodEvidence(prompt) && prompt.length > bestPlanLen) { bestPlanLen = prompt.length; bestPlanPrompt = prompt; }
       }
       if (exploratoryPatterns.test(prompt)) {
         exploratoryPrompts++;
-        if (prompt.length > bestExploreLen) { bestExploreLen = prompt.length; bestExplorePrompt = prompt; }
+        if (isGoodEvidence(prompt) && prompt.length > bestExploreLen) { bestExploreLen = prompt.length; bestExplorePrompt = prompt; }
       }
       if (boilerplatePatterns.test(prompt)) boilerplatePrompts++;
       if (testingPatterns.test(prompt)) testingPrompts++;

package/src/metrics/debug-cycles.js

@@ -10,6 +10,19 @@
  * - "it's still broken" vs targeted debug prompts
  */
 
+// ── Evidence quality filter ──
+// Prompts used as evidence should be human-written, readable, and illustrative.
+// Reject system-generated context, raw log pastes, and extreme lengths.
+const noisePatterns = /^This session is being continued|^\[?[0-9T:.Z-]{20,}|^\S+@\S+.*[%$#>]|^\s*\$\s|^\s*>\s/;
+function isGoodEvidence(prompt) {
+  if (!prompt || prompt.length < 40 || prompt.length > 600) return false;
+  if (noisePatterns.test(prompt)) return false;
+  // Reject if >40% of content is non-alpha (log lines, stack traces, JSON blobs)
+  const alpha = prompt.replace(/[^a-zA-Z]/g, '').length;
+  if (alpha / prompt.length < 0.4) return false;
+  return true;
+}
+
 const errorPatterns = /\b(error|bug|broken|crash|fail|exception|traceback|stack trace|doesn'?t work|not working|issue|problem|wrong)\b/i;
 const vaguePhrases = /^(it'?s? (?:still )?(?:not working|broken|wrong|failing))|^(fix it|try again|still (?:the same|broken|failing|not working))|^(same (?:error|issue|problem|thing))/i;
 const specificDebugPatterns = /\b(line \d+|TypeError|SyntaxError|ImportError|ReferenceError|ValueError|KeyError|AttributeError|NoneType|undefined is not|cannot read prop|stack trace|traceback|\.py:\d+|\.ts:\d+|\.js:\d+|status (?:code )?\d{3}|HTTP \d{3}|ENOENT|EACCES|CORS|404|500|502|503)\b/i;
@@ -57,8 +70,9 @@ export function computeDebugCycles(sessions) {
         }
         if (specificDebugPatterns.test(prompt) || prompt.length > 200) {
           specificReports++;
-          // Track best specific report
-          if (prompt.length > bestSpecificLen) {
+          // Track best specific report — require actual debug pattern match
+          // and readable evidence quality
+          if (specificDebugPatterns.test(prompt) && isGoodEvidence(prompt) && prompt.length > bestSpecificLen) {
             bestSpecificLen = prompt.length;
             bestSpecificReport = prompt;
           }
@@ -69,7 +83,7 @@ export function computeDebugCycles(sessions) {
           totalTurnsToResolve += debugTurnCount;
           if (debugTurnCount <= 2) {
             quickFixes++;
-            if (!bestQuickFix) bestQuickFix = debugStartPrompt;
+            if (!bestQuickFix && isGoodEvidence(debugStartPrompt)) bestQuickFix = debugStartPrompt;
           }
           if (debugTurnCount > 5) longLoops++;
           inDebugMode = false;

package/src/metrics/decomposition.js

@@ -11,6 +11,16 @@
  * - Follow-up prompts that reference or build on previous context
  */
 
+// ── Evidence quality filter ──
+const noisePatterns = /^This session is being continued|^\[?[0-9T:.Z-]{20,}|^\S+@\S+.*[%$#>]|^\s*\$\s|^\s*>\s/;
+function isGoodEvidence(prompt) {
+  if (!prompt || prompt.length < 40 || prompt.length > 600) return false;
+  if (noisePatterns.test(prompt)) return false;
+  const alpha = prompt.replace(/[^a-zA-Z]/g, '').length;
+  if (alpha / prompt.length < 0.4) return false;
+  return true;
+}
+
 export function computeDecomposition(sessions) {
   if (sessions.length === 0) return { score: 50, details: {} };
 
@@ -51,7 +61,7 @@ export function computeDecomposition(sessions) {
       if (len < 100) shortPromptCount++;
 
       // Track decomposition examples (multi-sentence prompts showing task breakdown)
-      if (len > 150 && len < 2000) {
+      if (isGoodEvidence(prompt)) {
         decompCandidates.push(prompt);
       }
 
@@ -60,7 +70,7 @@ export function computeDecomposition(sessions) {
         if (followupPatterns.test(prompt) || refinementPatterns.test(prompt)) {
           contextualFollowups++;
           // Capture best followup example
-          if (!bestFollowupPrompt || prompt.length > bestFollowupPrompt.length) {
+          if (isGoodEvidence(prompt) && (!bestFollowupPrompt || prompt.length > bestFollowupPrompt.length)) {
             bestFollowupPrompt = prompt;
           }
         }

package/src/metrics/session-structure.js

@@ -11,6 +11,16 @@
  * - Modification rate of AI output (shows critical review)
  */
 
+// ── Evidence quality filter ──
+const noisePatterns = /^This session is being continued|^\[?[0-9T:.Z-]{20,}|^\S+@\S+.*[%$#>]|^\s*\$\s|^\s*>\s/;
+function isGoodEvidence(prompt) {
+  if (!prompt || prompt.length < 40 || prompt.length > 600) return false;
+  if (noisePatterns.test(prompt)) return false;
+  const alpha = prompt.replace(/[^a-zA-Z]/g, '').length;
+  if (alpha / prompt.length < 0.4) return false;
+  return true;
+}
+
 const contextSettingPatterns = /^(i('?m| am) (working on|building|trying to|looking at)|we need to|the goal is|here'?s (the|what)|context:|background:|i have a|there'?s a|i want to|let me explain)/i;
 const planningStartPatterns = /^(let'?s (plan|think|figure|start by)|first,? (let'?s|we should)|before we (start|begin|code)|the plan is|step 1|here'?s (my|the) plan)/i;
 const reviewPatterns = /\b(looks good|ship it|deploy|push it|commit|merge|let'?s go|lgtm|approved|test it|run (the )?tests|build it|does this look|review this|check this)\b/i;
@@ -52,7 +62,7 @@ export function computeSessionStructure(sessions) {
     if (contextSettingPatterns.test(firstPrompt) || firstPrompt.length > 200) {
       contextSetSessions++;
       // Track best context-setting prompt
-      if (firstPrompt.length > bestContextLen) {
+      if (isGoodEvidence(firstPrompt) && firstPrompt.length > bestContextLen) {
         bestContextLen = firstPrompt.length;
         bestContextPrompt = firstPrompt;
       }
@@ -77,7 +87,7 @@ export function computeSessionStructure(sessions) {
       if (refinementPatterns.test(prompt)) {
         refinementCount++;
         // Track best refinement example
-        if (!bestRefinementPrompt || prompt.length > bestRefinementPrompt.length) {
+        if (isGoodEvidence(prompt) && (!bestRefinementPrompt || prompt.length > bestRefinementPrompt.length)) {
           bestRefinementPrompt = prompt;
         }
       }