@nerviq/cli 1.18.0 → 1.20.0

package/src/synergy/learning.js

@@ -1,199 +1,199 @@
-/**
- * S1. Cross-Platform Learning Engine
- *
- * When a recommendation succeeds on one platform, propagate to others.
- * Translates recommendations into each platform's native semantics
- * rather than blindly copying.
- */
-
-const fs = require('fs');
-const path = require('path');
-
-const PLATFORM_SEMANTICS = {
-  claude: { configFile: 'CLAUDE.md', format: 'markdown', instructionStyle: 'prose' },
-  codex: { configFile: 'AGENTS.md', format: 'markdown', instructionStyle: 'directive' },
-  gemini: { configFile: 'GEMINI.md', format: 'markdown', instructionStyle: 'structured' },
-  copilot: { configFile: '.github/copilot-instructions.md', format: 'markdown', instructionStyle: 'concise' },
-  cursor: { configFile: '.cursor/rules', format: 'mdc', instructionStyle: 'rule-based' },
-  windsurf: { configFile: '.windsurf/rules', format: 'markdown', instructionStyle: 'rule-based' },
-  aider: { configFile: '.aider.conf.yml', format: 'yaml', instructionStyle: 'convention-based' },
-  opencode: { configFile: 'opencode.json', format: 'jsonc', instructionStyle: 'directive' },
-};
-
-const TRANSLATION_MAP = {
-  'verification-command': {
-    claude: (rec) => `Add to CLAUDE.md: \`## Verification\` section with \`${rec.command}\``,
-    codex: (rec) => `Add to AGENTS.md verification block: \`${rec.command}\``,
-    gemini: (rec) => `Add to GEMINI.md: verification command \`${rec.command}\``,
-    copilot: (rec) => `Document in copilot-instructions.md: always run \`${rec.command}\` before completing`,
-    cursor: (rec) => `Add rule: "Always run \`${rec.command}\` before completing tasks"`,
-    windsurf: (rec) => `Add a .windsurf/rules verification rule that runs \`${rec.command}\` before task completion`,
-    aider: (rec) => `Document in CONVENTIONS.md or .aider.conf.yml workflow notes: run \`${rec.command}\` before final response`,
-    opencode: (rec) => `Add to AGENTS.md or opencode.json workflow guidance: run \`${rec.command}\` before completing`,
-  },
-  'architecture-doc': {
-    claude: () => `Add Mermaid architecture diagram to CLAUDE.md`,
-    codex: () => `Add architecture overview to AGENTS.md`,
-    gemini: () => `Add architecture context to GEMINI.md`,
-    copilot: () => `Add architecture reference in copilot-instructions.md`,
-    cursor: () => `Add architecture context rule in .cursor/rules`,
-    windsurf: () => `Add architecture guidance to a .windsurf/rules/*.md file or shared workflow`,
-    aider: () => `Add architecture overview to CONVENTIONS.md for Aider sessions`,
-    opencode: () => `Add architecture context to AGENTS.md and reference it from opencode.json`,
-  },
-  'permission-rule': {
-    claude: (rec) => `Configure in settings.json permissions: ${rec.description}`,
-    codex: (rec) => `Set sandbox policy in codex config: ${rec.description}`,
-    gemini: (rec) => `Configure Gemini sandbox settings: ${rec.description}`,
-    copilot: (rec) => `Note in instructions: ${rec.description}`,
-    cursor: (rec) => `Add safety rule: ${rec.description}`,
-    windsurf: (rec) => `Translate into .cascadeignore, MCP whitelist, or rule guidance: ${rec.description}`,
-    aider: (rec) => `Translate into git workflow guardrails or config defaults: ${rec.description}`,
-    opencode: (rec) => `Encode in opencode.json permissions: ${rec.description}`,
-  },
-  'hook': {
-    claude: (rec) => `Add hook in settings.json: ${rec.description}`,
-    codex: (rec) => `Add pre/post task check: ${rec.description}`,
-    gemini: (rec) => `Add automated check: ${rec.description}`,
-    copilot: (rec) => `Add workflow step: ${rec.description}`,
-    cursor: (rec) => `Add background task rule: ${rec.description}`,
-    windsurf: (rec) => `Implement as a workflow, Steps sequence, or auto rule: ${rec.description}`,
-    aider: (rec) => `Implement via git hooks, auto-test, or documented Aider workflow: ${rec.description}`,
-    opencode: (rec) => `Implement via plugin, workflow, or task automation: ${rec.description}`,
-  },
-};
-
-/**
- * Adapt a recommendation from one platform's semantics to another.
- */
-function adaptRecommendation(recommendation, sourcePlatform, targetPlatform) {
-  const translator = TRANSLATION_MAP[recommendation.type];
-  if (translator && translator[targetPlatform]) {
-    return translator[targetPlatform](recommendation);
-  }
-  // Generic fallback: describe the recommendation in platform-neutral terms
-  const semantics = PLATFORM_SEMANTICS[targetPlatform];
-  if (!semantics) return recommendation.description || recommendation.recommendation;
-  return `[${semantics.configFile}] ${recommendation.description || recommendation.recommendation}`;
-}
-
-/**
- * Calculate confidence for a propagated insight based on similarity
- * between source and target platforms.
- */
-function calculatePropagationConfidence(insight, sourcePlatform, targetPlatform) {
-  // Base confidence from the original outcome
-  let confidence = Math.min(1, Math.max(0, (insight.score_delta || 0) / 20));
-
-  // Same instruction style = higher confidence
-  const sourceStyle = (PLATFORM_SEMANTICS[sourcePlatform] || {}).instructionStyle;
-  const targetStyle = (PLATFORM_SEMANTICS[targetPlatform] || {}).instructionStyle;
-  if (sourceStyle === targetStyle) confidence *= 1.2;
-
-  // Same format = slightly higher confidence
-  const sourceFormat = (PLATFORM_SEMANTICS[sourcePlatform] || {}).format;
-  const targetFormat = (PLATFORM_SEMANTICS[targetPlatform] || {}).format;
-  if (sourceFormat === targetFormat) confidence *= 1.1;
-
-  return Math.min(1, Math.round(confidence * 100) / 100);
-}
-
-/**
- * Propagate an insight from one platform to others.
- *
- * @param {Object} insight - { type, recommendation, outcome, score_delta }
- * @param {string} sourcePlatform - Platform where the insight originated
- * @param {string[]} targetPlatforms - Platforms to propagate to
- * @returns {Object[]} Adapted recommendations per target platform
- */
-function propagateInsight(insight, sourcePlatform, targetPlatforms) {
-  return targetPlatforms
-    .filter(p => p !== sourcePlatform)
-    .map(platform => ({
-      platform,
-      adaptedRecommendation: adaptRecommendation(insight, sourcePlatform, platform),
-      confidence: calculatePropagationConfidence(insight, sourcePlatform, platform),
-      sourceEvidence: {
-        platform: sourcePlatform,
-        outcome: insight.outcome,
-        score_delta: insight.score_delta,
-      },
-    }));
-}
-
-/**
- * Read recommendation outcomes from all platforms and find cross-learning opportunities.
- *
- * @param {string} dir - Project directory
- * @returns {Object[]} Cross-learning opportunities
- */
-function getCrossLearnings(dir) {
-  const learningsFile = path.join(dir, '.claude', 'synergy-learnings.json');
-  let learnings = [];
-  try {
-    learnings = JSON.parse(fs.readFileSync(learningsFile, 'utf8'));
-  } catch {
-    // No learnings file yet
-  }
-
-  // Group by recommendation type
-  const byType = {};
-  for (const learning of learnings) {
-    if (!byType[learning.type]) byType[learning.type] = [];
-    byType[learning.type].push(learning);
-  }
-
-  const opportunities = [];
-  for (const [type, items] of Object.entries(byType)) {
-    // Find successful outcomes
-    const successes = items.filter(i => i.outcome === 'applied' || i.outcome === 'helpful');
-    if (successes.length === 0) continue;
-
-    // Find platforms that haven't tried this yet
-    const triedPlatforms = new Set(items.map(i => i.platform));
-    const allPlatforms = Object.keys(PLATFORM_SEMANTICS);
-    const untried = allPlatforms.filter(p => !triedPlatforms.has(p));
-
-    if (untried.length > 0) {
-      const bestSuccess = successes.reduce((a, b) =>
-        (b.score_delta || 0) > (a.score_delta || 0) ? b : a
-      );
-      opportunities.push({
-        type,
-        successfulOn: successes.map(s => s.platform),
-        untestedOn: untried,
-        bestEvidence: bestSuccess,
-        propagations: propagateInsight(bestSuccess, bestSuccess.platform, untried),
-      });
-    }
-  }
-
-  return opportunities;
-}
-
-/**
- * Record a learning outcome for later cross-platform propagation.
- *
- * @param {string} dir - Project directory
- * @param {Object} outcome - { platform, type, recommendation, outcome, score_delta }
- */
-function recordLearningOutcome(dir, outcome) {
-  const learningsFile = path.join(dir, '.claude', 'synergy-learnings.json');
-  let learnings = [];
-  try {
-    learnings = JSON.parse(fs.readFileSync(learningsFile, 'utf8'));
-  } catch {
-    // Start fresh
-  }
-
-  learnings.push({
-    ...outcome,
-    timestamp: new Date().toISOString(),
-  });
-
-  const claudeDir = path.join(dir, '.claude');
-  if (!fs.existsSync(claudeDir)) fs.mkdirSync(claudeDir, { recursive: true });
-  fs.writeFileSync(learningsFile, JSON.stringify(learnings, null, 2));
-}
-
-module.exports = { propagateInsight, getCrossLearnings, recordLearningOutcome };
+/**
+ * S1. Cross-Platform Learning Engine
+ *
+ * When a recommendation succeeds on one platform, propagate to others.
+ * Translates recommendations into each platform's native semantics
+ * rather than blindly copying.
+ */
+
+const fs = require('fs');
+const path = require('path');
+
+const PLATFORM_SEMANTICS = {
+  claude: { configFile: 'CLAUDE.md', format: 'markdown', instructionStyle: 'prose' },
+  codex: { configFile: 'AGENTS.md', format: 'markdown', instructionStyle: 'directive' },
+  gemini: { configFile: 'GEMINI.md', format: 'markdown', instructionStyle: 'structured' },
+  copilot: { configFile: '.github/copilot-instructions.md', format: 'markdown', instructionStyle: 'concise' },
+  cursor: { configFile: '.cursor/rules', format: 'mdc', instructionStyle: 'rule-based' },
+  windsurf: { configFile: '.windsurf/rules', format: 'markdown', instructionStyle: 'rule-based' },
+  aider: { configFile: '.aider.conf.yml', format: 'yaml', instructionStyle: 'convention-based' },
+  opencode: { configFile: 'opencode.json', format: 'jsonc', instructionStyle: 'directive' },
+};
+
+const TRANSLATION_MAP = {
+  'verification-command': {
+    claude: (rec) => `Add to CLAUDE.md: \`## Verification\` section with \`${rec.command}\``,
+    codex: (rec) => `Add to AGENTS.md verification block: \`${rec.command}\``,
+    gemini: (rec) => `Add to GEMINI.md: verification command \`${rec.command}\``,
+    copilot: (rec) => `Document in copilot-instructions.md: always run \`${rec.command}\` before completing`,
+    cursor: (rec) => `Add rule: "Always run \`${rec.command}\` before completing tasks"`,
+    windsurf: (rec) => `Add a .windsurf/rules verification rule that runs \`${rec.command}\` before task completion`,
+    aider: (rec) => `Document in CONVENTIONS.md or .aider.conf.yml workflow notes: run \`${rec.command}\` before final response`,
+    opencode: (rec) => `Add to AGENTS.md or opencode.json workflow guidance: run \`${rec.command}\` before completing`,
+  },
+  'architecture-doc': {
+    claude: () => `Add Mermaid architecture diagram to CLAUDE.md`,
+    codex: () => `Add architecture overview to AGENTS.md`,
+    gemini: () => `Add architecture context to GEMINI.md`,
+    copilot: () => `Add architecture reference in copilot-instructions.md`,
+    cursor: () => `Add architecture context rule in .cursor/rules`,
+    windsurf: () => `Add architecture guidance to a .windsurf/rules/*.md file or shared workflow`,
+    aider: () => `Add architecture overview to CONVENTIONS.md for Aider sessions`,
+    opencode: () => `Add architecture context to AGENTS.md and reference it from opencode.json`,
+  },
+  'permission-rule': {
+    claude: (rec) => `Configure in settings.json permissions: ${rec.description}`,
+    codex: (rec) => `Set sandbox policy in codex config: ${rec.description}`,
+    gemini: (rec) => `Configure Gemini sandbox settings: ${rec.description}`,
+    copilot: (rec) => `Note in instructions: ${rec.description}`,
+    cursor: (rec) => `Add safety rule: ${rec.description}`,
+    windsurf: (rec) => `Translate into .cascadeignore, MCP whitelist, or rule guidance: ${rec.description}`,
+    aider: (rec) => `Translate into git workflow guardrails or config defaults: ${rec.description}`,
+    opencode: (rec) => `Encode in opencode.json permissions: ${rec.description}`,
+  },
+  'hook': {
+    claude: (rec) => `Add hook in settings.json: ${rec.description}`,
+    codex: (rec) => `Add pre/post task check: ${rec.description}`,
+    gemini: (rec) => `Add automated check: ${rec.description}`,
+    copilot: (rec) => `Add workflow step: ${rec.description}`,
+    cursor: (rec) => `Add background task rule: ${rec.description}`,
+    windsurf: (rec) => `Implement as a workflow, Steps sequence, or auto rule: ${rec.description}`,
+    aider: (rec) => `Implement via git hooks, auto-test, or documented Aider workflow: ${rec.description}`,
+    opencode: (rec) => `Implement via plugin, workflow, or task automation: ${rec.description}`,
+  },
+};
+
+/**
+ * Adapt a recommendation from one platform's semantics to another.
+ */
+function adaptRecommendation(recommendation, sourcePlatform, targetPlatform) {
+  const translator = TRANSLATION_MAP[recommendation.type];
+  if (translator && translator[targetPlatform]) {
+    return translator[targetPlatform](recommendation);
+  }
+  // Generic fallback: describe the recommendation in platform-neutral terms
+  const semantics = PLATFORM_SEMANTICS[targetPlatform];
+  if (!semantics) return recommendation.description || recommendation.recommendation;
+  return `[${semantics.configFile}] ${recommendation.description || recommendation.recommendation}`;
+}
+
+/**
+ * Calculate confidence for a propagated insight based on similarity
+ * between source and target platforms.
+ */
+function calculatePropagationConfidence(insight, sourcePlatform, targetPlatform) {
+  // Base confidence from the original outcome
+  let confidence = Math.min(1, Math.max(0, (insight.score_delta || 0) / 20));
+
+  // Same instruction style = higher confidence
+  const sourceStyle = (PLATFORM_SEMANTICS[sourcePlatform] || {}).instructionStyle;
+  const targetStyle = (PLATFORM_SEMANTICS[targetPlatform] || {}).instructionStyle;
+  if (sourceStyle === targetStyle) confidence *= 1.2;
+
+  // Same format = slightly higher confidence
+  const sourceFormat = (PLATFORM_SEMANTICS[sourcePlatform] || {}).format;
+  const targetFormat = (PLATFORM_SEMANTICS[targetPlatform] || {}).format;
+  if (sourceFormat === targetFormat) confidence *= 1.1;
+
+  return Math.min(1, Math.round(confidence * 100) / 100);
+}
+
+/**
+ * Propagate an insight from one platform to others.
+ *
+ * @param {Object} insight - { type, recommendation, outcome, score_delta }
+ * @param {string} sourcePlatform - Platform where the insight originated
+ * @param {string[]} targetPlatforms - Platforms to propagate to
+ * @returns {Object[]} Adapted recommendations per target platform
+ */
+function propagateInsight(insight, sourcePlatform, targetPlatforms) {
+  return targetPlatforms
+    .filter(p => p !== sourcePlatform)
+    .map(platform => ({
+      platform,
+      adaptedRecommendation: adaptRecommendation(insight, sourcePlatform, platform),
+      confidence: calculatePropagationConfidence(insight, sourcePlatform, platform),
+      sourceEvidence: {
+        platform: sourcePlatform,
+        outcome: insight.outcome,
+        score_delta: insight.score_delta,
+      },
+    }));
+}
+
+/**
+ * Read recommendation outcomes from all platforms and find cross-learning opportunities.
+ *
+ * @param {string} dir - Project directory
+ * @returns {Object[]} Cross-learning opportunities
+ */
+function getCrossLearnings(dir) {
+  const learningsFile = path.join(dir, '.claude', 'synergy-learnings.json');
+  let learnings = [];
+  try {
+    learnings = JSON.parse(fs.readFileSync(learningsFile, 'utf8'));
+  } catch {
+    // No learnings file yet
+  }
+
+  // Group by recommendation type
+  const byType = {};
+  for (const learning of learnings) {
+    if (!byType[learning.type]) byType[learning.type] = [];
+    byType[learning.type].push(learning);
+  }
+
+  const opportunities = [];
+  for (const [type, items] of Object.entries(byType)) {
+    // Find successful outcomes
+    const successes = items.filter(i => i.outcome === 'applied' || i.outcome === 'helpful');
+    if (successes.length === 0) continue;
+
+    // Find platforms that haven't tried this yet
+    const triedPlatforms = new Set(items.map(i => i.platform));
+    const allPlatforms = Object.keys(PLATFORM_SEMANTICS);
+    const untried = allPlatforms.filter(p => !triedPlatforms.has(p));
+
+    if (untried.length > 0) {
+      const bestSuccess = successes.reduce((a, b) =>
+        (b.score_delta || 0) > (a.score_delta || 0) ? b : a
+      );
+      opportunities.push({
+        type,
+        successfulOn: successes.map(s => s.platform),
+        untestedOn: untried,
+        bestEvidence: bestSuccess,
+        propagations: propagateInsight(bestSuccess, bestSuccess.platform, untried),
+      });
+    }
+  }
+
+  return opportunities;
+}
+
+/**
+ * Record a learning outcome for later cross-platform propagation.
+ *
+ * @param {string} dir - Project directory
+ * @param {Object} outcome - { platform, type, recommendation, outcome, score_delta }
+ */
+function recordLearningOutcome(dir, outcome) {
+  const learningsFile = path.join(dir, '.claude', 'synergy-learnings.json');
+  let learnings = [];
+  try {
+    learnings = JSON.parse(fs.readFileSync(learningsFile, 'utf8'));
+  } catch {
+    // Start fresh
+  }
+
+  learnings.push({
+    ...outcome,
+    timestamp: new Date().toISOString(),
+  });
+
+  const claudeDir = path.join(dir, '.claude');
+  if (!fs.existsSync(claudeDir)) fs.mkdirSync(claudeDir, { recursive: true });
+  fs.writeFileSync(learningsFile, JSON.stringify(learnings, null, 2));
+}
+
+module.exports = { propagateInsight, getCrossLearnings, recordLearningOutcome };