tribunal-kit 9.2.4 → 9.2.5

package/{dist → src}/commands/optimize.js

@@ -1,328 +1,332 @@
-"use strict";
-var __importDefault = (this && this.__importDefault) || function (mod) {
-    return (mod && mod.__esModule) ? mod : { "default": mod };
-};
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.cmdOptimizeSkill = cmdOptimizeSkill;
-
-const fs = __importDefault(require("fs"));
-const path = __importDefault(require("path"));
-const os = __importDefault(require("os"));
-const https = __importDefault(require("https"));
-const { execSync, spawnSync } = require("child_process");
-const logger = require("../utils/logger");
-const helpers = require("../utils/helpers");
-
-// ── LLM Helpers ──────────────────────────────────────────────────────────────
-function detectLlmProvider() {
-    if (process.env.ANTHROPIC_API_KEY)
-        return { provider: "anthropic", key: process.env.ANTHROPIC_API_KEY };
-    if (process.env.OPENAI_API_KEY)
-        return { provider: "openai", key: process.env.OPENAI_API_KEY };
-    if (process.env.GEMINI_API_KEY)
-        return { provider: "gemini", key: process.env.GEMINI_API_KEY };
-    return null;
-}
-
-function httpsPost(hostname, apiPath, headers, body) {
-    return new Promise((resolve, reject) => {
-        const data = JSON.stringify(body);
-        const req = https.request(
-            {
-                method: "POST",
-                hostname,
-                path: apiPath,
-                headers: { ...headers, "Content-Length": Buffer.byteLength(data) },
-            },
-            (res) => {
-                let raw = "";
-                res.on("data", (c) => { raw += c; });
-                res.on("end", () => resolve(raw));
-                res.on("error", reject);
-            },
-        );
-        req.on("error", reject);
-        req.setTimeout(30000, () => {
-            req.destroy(new Error("LLM API timeout"));
-        });
-        req.write(data);
-        req.end();
-    });
-}
-
-async function callLlmApi(prompt, provider, apiKey) {
-    try {
-        if (provider === "anthropic") {
-            const raw = await httpsPost(
-                "api.anthropic.com",
-                "/v1/messages",
-                {
-                    "Content-Type": "application/json",
-                    "x-api-key": apiKey,
-                    "anthropic-version": "2023-06-01",
-                },
-                {
-                    model: "claude-3-5-sonnet-latest",
-                    max_tokens: 1024,
-                    messages: [{ role: "user", content: prompt }],
-                },
-            );
-            const json = JSON.parse(raw);
-            return json?.content?.[0]?.text ?? null;
-        }
-
-        if (provider === "openai") {
-            const raw = await httpsPost(
-                "api.openai.com",
-                "/v1/chat/completions",
-                {
-                    "Content-Type": "application/json",
-                    Authorization: `Bearer ${apiKey}`,
-                },
-                {
-                    model: "gpt-4o-mini",
-                    max_tokens: 1024,
-                    messages: [{ role: "user", content: prompt }],
-                    temperature: 0.1,
-                },
-            );
-            const json = JSON.parse(raw);
-            return json?.choices?.[0]?.message?.content ?? null;
-        }
-
-        if (provider === "gemini") {
-            const raw = await httpsPost(
-                "generativelanguage.googleapis.com",
-                `/v1beta/models/gemini-1.5-flash:generateContent?key=${apiKey}`,
-                { "Content-Type": "application/json" },
-                {
-                    contents: [{ parts: [{ text: prompt }] }],
-                    generationConfig: { maxOutputTokens: 1024, temperature: 0.1 },
-                },
-            );
-            const json = JSON.parse(raw);
-            return json?.candidates?.[0]?.content?.parts?.[0]?.text ?? null;
-        }
-    } catch (e) {
-        logger.err(`LLM API Call failed: ${e.message}`);
-        return null;
-    }
-    return null;
-}
-
-// ── Core Binary Resolver ─────────────────────────────────────────────────────
-function getCoreBinaryPath(targetDir) {
-    const isWindows = os.default.platform() === "win32";
-    const ext = isWindows ? ".exe" : "";
-    
-    // Dev build target checks
-    const devPath = path.default.resolve(targetDir, 'target', 'release', `tribunal-core${ext}`);
-    if (fs.default.existsSync(devPath)) return devPath;
-
-    const debugPath = path.default.resolve(targetDir, 'target', 'debug', `tribunal-core${ext}`);
-    if (fs.default.existsSync(debugPath)) return debugPath;
-
-    // Production npm package checks
-    const pkgName = `@tribunal-kit/core-${os.default.platform()}-${os.default.arch()}`;
-    try {
-        const pkgPath = require.resolve(`${pkgName}/package.json`);
-        const binPath = path.default.resolve(path.default.dirname(pkgPath), `bin/tribunal-core${ext}`);
-        if (fs.default.existsSync(binPath)) return binPath;
-    } catch {}
-
-    return null;
-}
-
-// ── Main Optimizer Command ───────────────────────────────────────────────────
-async function cmdOptimizeSkill(flags, processArgs, quiet = false) {
-    const targetDir = flags.path ? path.default.resolve(flags.path) : process.cwd();
-    const agentDest = path.default.join(targetDir, '.agent');
-
-    if (!fs.default.existsSync(agentDest)) {
-        logger.err('.agent/ not found. Run: npx tribunal-kit init');
-        process.exit(1);
-    }
-
-    helpers.banner(quiet);
-
-    // Detect LLM Credentials
-    const llm = detectLlmProvider();
-    if (!llm) {
-        logger.err("No LLM API Key detected in environment. Please set GEMINI_API_KEY, OPENAI_API_KEY, or ANTHROPIC_API_KEY.");
-        process.exit(1);
-    }
-
-    // Resolve Core Binary
-    const binPath = getCoreBinaryPath(targetDir);
-    if (!binPath) {
-        logger.err("tribunal-core binary not found. Run 'cargo build --release' or install the core package.");
-        process.exit(1);
-    }
-
-    // Parse loop parameters
-    const skillName = flags.target || 'project-idioms';
-    const skillPath = path.default.join(agentDest, 'skills', skillName, 'SKILL.md');
-    
-    if (!fs.default.existsSync(skillPath)) {
-        logger.err(`Target skill file not found: ${skillPath}`);
-        process.exit(1);
-    }
-
-    // Find the harness command
-    const positionalArgs = processArgs.slice(3).filter(arg => !arg.startsWith("--"));
-    if (positionalArgs.length === 0) {
-        logger.err("Harness command is required. Usage: tk optimize-skill --target <skill-name> \"<harness-command>\"");
-        process.exit(1);
-    }
-    const harnessCmd = positionalArgs[0];
-
-    const epochs = 4;
-    const editBudget = 4;
-
-    logger.log(`  ${logger.c('cyan', '⊘')} Starting SkillOpt loop for skill: ${logger.bold(skillName)}`);
-    logger.log(`  ${logger.c('cyan', '⊘')} Harness: "${logger.bold(harnessCmd)}"`);
-    logger.log(`  ${logger.c('cyan', '⊘')} Provider: ${logger.bold(llm.provider)}`);
-    console.log();
-
-    let bestScore = 0.0;
-    let currentSkillText = fs.default.readFileSync(skillPath, 'utf8');
-
-    for (let epoch = 1; epoch <= epochs; epoch++) {
-        logger.log(`  ${logger.c('cyan', '▶')} ${logger.bold(`Epoch ${epoch}/${epochs}`)}`);
-
-        // 1. Run Rollout Harness
-        logger.log(`    ${logger.c('gray', '●')} Running rollout batch execution...`);
-        let rolloutLogs = "";
-        let rolloutSuccess = true;
-        try {
-            rolloutLogs = execSync(harnessCmd, { encoding: 'utf8', cwd: targetDir, stdio: 'pipe' });
-        } catch (e) {
-            rolloutLogs = e.stdout + "\n" + e.stderr;
-            rolloutSuccess = false;
-        }
-
-        const _initialScore = rolloutSuccess ? 1.0 : 0.0;
-        logger.log(`    ${logger.c('gray', '●')} Rollout score: ${rolloutSuccess ? logger.c('green', '1.0') : logger.c('red', '0.0')}`);
-
-        // 2. Call LLM for Patch Proposals
-        logger.log(`    ${logger.c('gray', '●')} Requesting optimizer reflections from LLM...`);
-        const reflectionPrompt = `
-You are a SkillOpt text-space optimizer coach.
-Analyze the following rollout run logs and current skill document, then propose bounded add/delete/replace/insert_after edits.
-The goal is to fix failure patterns and make the procedural memory rules clearer and more robust.
-
-Current Skill Content:
-"""
-${currentSkillText}
-"""
-
-Rollout Logs:
-${rolloutLogs.slice(-4000)}
-
-Output ONLY a JSON array containing proposed edits. No markdown formatting, no explanation.
-Each edit must follow this schema:
-[
-  {
-    "op": "append" | "insert_after" | "replace" | "delete",
-    "target": "<exact text match to replace/insert_after/delete, leave null for append>",
-    "content": "<exact markdown/rules to add, replace, or insert>",
-    "support_count": 1,
-    "source_type": "${rolloutSuccess ? "success" : "failure"}"
-  }
-]
-`;
-        
-        const llmResponse = await callLlmApi(reflectionPrompt, llm.provider, llm.key);
-        if (!llmResponse) {
-            logger.warn("    ⚠️ LLM returned empty response. Skipping epoch step.");
-            continue;
-        }
-
-        // Clean LLM JSON response from markdown wrappers if present
-        let cleanedJson = llmResponse.trim();
-        if (cleanedJson.startsWith("```json")) {
-            cleanedJson = cleanedJson.replace(/^```json/, "").replace(/```$/, "").trim();
-        } else if (cleanedJson.startsWith("```")) {
-            cleanedJson = cleanedJson.replace(/^```/, "").replace(/```$/, "").trim();
-        }
-
-        // 3. Invoke Rust binary for optimizing step
-        logger.log(`    ${logger.c('gray', '●')} Merging and applying bounded edits via Rust Core...`);
-        let optResultStr = "";
-        try {
-            const tempFile = path.default.join(os.default.tmpdir(), `edits-${Date.now()}.json`);
-            fs.default.writeFileSync(tempFile, cleanedJson, 'utf8');
-            
-            const result = spawnSync(binPath, [
-                "optimize-step",
-                "--skill-path",
-                skillPath,
-                "--edits-json",
-                cleanedJson,
-                "--budget",
-                String(editBudget)
-            ], { encoding: 'utf8', cwd: targetDir, stdio: 'pipe' });
-            
-            if (result.error) {
-                throw result.error;
-            }
-            optResultStr = result.stdout || "";
-            
-            try { fs.default.unlinkSync(tempFile); } catch {}
-        } catch (e) {
-            logger.err(`    ✖ Rust Core optimization failed: ${e.message}`);
-            continue;
-        }
-
-        let optReport;
-        try {
-            optReport = JSON.parse(optResultStr.trim());
-        } catch {
-            logger.err("    ✖ Failed to parse optimizer report from Rust Core.");
-            continue;
-        }
-
-        if (optReport.applied_count === 0) {
-            logger.log(`    ${logger.c('yellow', '●')} No edits applied (deduplicated or target not found).`);
-            continue;
-        }
-
-        logger.log(`    ${logger.c('green', '●')} Applied ${optReport.applied_count} edits to skill.`);
-
-        // 4. Run Validation Harness
-        logger.log(`    ${logger.c('gray', '●')} Evaluating candidate skill against validation split...`);
-        let valSuccess = true;
-        try {
-            execSync(harnessCmd, { encoding: 'utf8', cwd: targetDir, stdio: 'pipe' });
-        } catch {
-            valSuccess = false;
-        }
-
-        const candidateScore = valSuccess ? 1.0 : 0.0;
-
-        // 5. Validation Gate
-        if (candidateScore > bestScore || (candidateScore === 1.0 && bestScore === 0.0)) {
-            logger.ok(`    ✔ Gate Accepted! Score improved/stabilized to ${logger.c('green', String(candidateScore))}.`);
-            bestScore = candidateScore;
-            currentSkillText = fs.default.readFileSync(skillPath, 'utf8');
-        } else {
-            logger.warn(`    ⚠️ Gate Rejected. Score (${candidateScore}) <= Best (${bestScore}). Reverting edits.`);
-            fs.default.writeFileSync(skillPath, currentSkillText, 'utf8');
-            
-            // Save to rejected-edit buffer history
-            const rejectedPath = path.default.join(agentDest, 'history', 'skill-optimization');
-            fs.default.mkdirSync(rejectedPath, { recursive: true });
-            const bufFile = path.default.join(rejectedPath, 'rejected-buffer.json');
-            let buffer = [];
-            if (fs.default.existsSync(bufFile)) {
-                try { buffer = JSON.parse(fs.default.readFileSync(bufFile, 'utf8')); } catch {}
-            }
-            buffer.push({ epoch, timestamp: new Date().toISOString(), edits: optReport.reports });
-            fs.default.writeFileSync(bufFile, JSON.stringify(buffer, null, 2), 'utf8');
-        }
-        console.log();
-    }
-
-    logger.ok(`🎉 SkillOpt training complete. Best validation score: ${logger.c('green', String(bestScore))}`);
-    console.log();
-}
+"use strict";
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.cmdOptimizeSkill = cmdOptimizeSkill;
+
+const fs = __importDefault(require("fs"));
+const path = __importDefault(require("path"));
+const os = __importDefault(require("os"));
+const https = __importDefault(require("https"));
+const { execFileSync, spawnSync } = require("child_process");
+const logger = require("../utils/logger");
+const helpers = require("../utils/helpers");
+
+// ── LLM Helpers ──────────────────────────────────────────────────────────────
+function detectLlmProvider() {
+    if (process.env.ANTHROPIC_API_KEY)
+        return { provider: "anthropic", key: process.env.ANTHROPIC_API_KEY };
+    if (process.env.OPENAI_API_KEY)
+        return { provider: "openai", key: process.env.OPENAI_API_KEY };
+    if (process.env.GEMINI_API_KEY)
+        return { provider: "gemini", key: process.env.GEMINI_API_KEY };
+    return null;
+}
+
+function httpsPost(hostname, apiPath, headers, body) {
+    return new Promise((resolve, reject) => {
+        const data = JSON.stringify(body);
+        const req = https.request(
+            {
+                method: "POST",
+                hostname,
+                path: apiPath,
+                headers: { ...headers, "Content-Length": Buffer.byteLength(data) },
+            },
+            (res) => {
+                let raw = "";
+                res.on("data", (c) => { raw += c; });
+                res.on("end", () => resolve(raw));
+                res.on("error", reject);
+            },
+        );
+        req.on("error", reject);
+        req.setTimeout(30000, () => {
+            req.destroy(new Error("LLM API timeout"));
+        });
+        req.write(data);
+        req.end();
+    });
+}
+
+async function callLlmApi(prompt, provider, apiKey) {
+    try {
+        if (provider === "anthropic") {
+            const raw = await httpsPost(
+                "api.anthropic.com",
+                "/v1/messages",
+                {
+                    "Content-Type": "application/json",
+                    "x-api-key": apiKey,
+                    "anthropic-version": "2023-06-01",
+                },
+                {
+                    model: "claude-3-5-sonnet-latest",
+                    max_tokens: 1024,
+                    messages: [{ role: "user", content: prompt }],
+                },
+            );
+            const json = JSON.parse(raw);
+            return json?.content?.[0]?.text ?? null;
+        }
+
+        if (provider === "openai") {
+            const raw = await httpsPost(
+                "api.openai.com",
+                "/v1/chat/completions",
+                {
+                    "Content-Type": "application/json",
+                    Authorization: `Bearer ${apiKey}`,
+                },
+                {
+                    model: "gpt-4o-mini",
+                    max_tokens: 1024,
+                    messages: [{ role: "user", content: prompt }],
+                    temperature: 0.1,
+                },
+            );
+            const json = JSON.parse(raw);
+            return json?.choices?.[0]?.message?.content ?? null;
+        }
+
+        if (provider === "gemini") {
+            const raw = await httpsPost(
+                "generativelanguage.googleapis.com",
+                `/v1beta/models/gemini-1.5-flash:generateContent?key=${apiKey}`,
+                { "Content-Type": "application/json" },
+                {
+                    contents: [{ parts: [{ text: prompt }] }],
+                    generationConfig: { maxOutputTokens: 1024, temperature: 0.1 },
+                },
+            );
+            const json = JSON.parse(raw);
+            return json?.candidates?.[0]?.content?.parts?.[0]?.text ?? null;
+        }
+    } catch (e) {
+        logger.err(`LLM API Call failed: ${e.message}`);
+        return null;
+    }
+    return null;
+}
+
+// ── Core Binary Resolver ─────────────────────────────────────────────────────
+function getCoreBinaryPath(targetDir) {
+    const isWindows = os.default.platform() === "win32";
+    const ext = isWindows ? ".exe" : "";
+    
+    // Dev build target checks
+    const devPath = path.default.resolve(targetDir, 'target', 'release', `tribunal-core${ext}`);
+    if (fs.default.existsSync(devPath)) return devPath;
+
+    const debugPath = path.default.resolve(targetDir, 'target', 'debug', `tribunal-core${ext}`);
+    if (fs.default.existsSync(debugPath)) return debugPath;
+
+    // Production npm package checks
+    const pkgName = `@tribunal-kit/core-${os.default.platform()}-${os.default.arch()}`;
+    try {
+        const pkgPath = require.resolve(`${pkgName}/package.json`);
+        const binPath = path.default.resolve(path.default.dirname(pkgPath), `bin/tribunal-core${ext}`);
+        if (fs.default.existsSync(binPath)) return binPath;
+    } catch {}
+
+    return null;
+}
+
+// ── Main Optimizer Command ───────────────────────────────────────────────────
+async function cmdOptimizeSkill(flags, processArgs, quiet = false) {
+    const targetDir = flags.path ? path.default.resolve(flags.path) : process.cwd();
+    const agentDest = path.default.join(targetDir, '.agent');
+
+    if (!fs.default.existsSync(agentDest)) {
+        logger.err('.agent/ not found. Run: npx tribunal-kit init');
+        process.exit(1);
+    }
+
+    helpers.banner(quiet);
+
+    // Detect LLM Credentials
+    const llm = detectLlmProvider();
+    if (!llm) {
+        logger.err("No LLM API Key detected in environment. Please set GEMINI_API_KEY, OPENAI_API_KEY, or ANTHROPIC_API_KEY.");
+        process.exit(1);
+    }
+
+    // Resolve Core Binary
+    const binPath = getCoreBinaryPath(targetDir);
+    if (!binPath) {
+        logger.err("tribunal-core binary not found. Run 'cargo build --release' or install the core package.");
+        process.exit(1);
+    }
+
+    // Parse loop parameters
+    const skillName = flags.target || 'project-idioms';
+    const skillPath = path.default.join(agentDest, 'skills', skillName, 'SKILL.md');
+    
+    if (!fs.default.existsSync(skillPath)) {
+        logger.err(`Target skill file not found: ${skillPath}`);
+        process.exit(1);
+    }
+
+    // Find the harness command
+    const positionalArgs = processArgs.slice(3).filter(arg => !arg.startsWith("--"));
+    if (positionalArgs.length === 0) {
+        logger.err("Harness command is required. Usage: tk optimize-skill --target <skill-name> \"<harness-command>\"");
+        process.exit(1);
+    }
+    const harnessCmd = positionalArgs[0];
+
+    const epochs = 4;
+    const editBudget = 4;
+
+    logger.log(`  ${logger.c('cyan', '⊘')} Starting SkillOpt loop for skill: ${logger.bold(skillName)}`);
+    logger.log(`  ${logger.c('cyan', '⊘')} Harness: "${logger.bold(harnessCmd)}"`);
+    logger.log(`  ${logger.c('cyan', '⊘')} Provider: ${logger.bold(llm.provider)}`);
+    console.log();
+
+    let bestScore = 0.0;
+    let currentSkillText = fs.default.readFileSync(skillPath, 'utf8');
+
+    for (let epoch = 1; epoch <= epochs; epoch++) {
+        logger.log(`  ${logger.c('cyan', '▶')} ${logger.bold(`Epoch ${epoch}/${epochs}`)}`);
+
+        // 1. Run Rollout Harness
+        logger.log(`    ${logger.c('gray', '●')} Running rollout batch execution...`);
+        let rolloutLogs = "";
+        let rolloutSuccess = true;
+        try {
+            const args = harnessCmd.trim().split(/\s+/);
+            const cmd = args.shift();
+            rolloutLogs = execFileSync(cmd, args, { encoding: 'utf8', cwd: targetDir, stdio: 'pipe' });
+        } catch (e) {
+            rolloutLogs = e.stdout + "\n" + e.stderr;
+            rolloutSuccess = false;
+        }
+
+        const _initialScore = rolloutSuccess ? 1.0 : 0.0;
+        logger.log(`    ${logger.c('gray', '●')} Rollout score: ${rolloutSuccess ? logger.c('green', '1.0') : logger.c('red', '0.0')}`);
+
+        // 2. Call LLM for Patch Proposals
+        logger.log(`    ${logger.c('gray', '●')} Requesting optimizer reflections from LLM...`);
+        const reflectionPrompt = `
+You are a SkillOpt text-space optimizer coach.
+Analyze the following rollout run logs and current skill document, then propose bounded add/delete/replace/insert_after edits.
+The goal is to fix failure patterns and make the procedural memory rules clearer and more robust.
+
+Current Skill Content:
+"""
+${currentSkillText}
+"""
+
+Rollout Logs:
+${rolloutLogs.slice(-4000)}
+
+Output ONLY a JSON array containing proposed edits. No markdown formatting, no explanation.
+Each edit must follow this schema:
+[
+  {
+    "op": "append" | "insert_after" | "replace" | "delete",
+    "target": "<exact text match to replace/insert_after/delete, leave null for append>",
+    "content": "<exact markdown/rules to add, replace, or insert>",
+    "support_count": 1,
+    "source_type": "${rolloutSuccess ? "success" : "failure"}"
+  }
+]
+`;
+        
+        const llmResponse = await callLlmApi(reflectionPrompt, llm.provider, llm.key);
+        if (!llmResponse) {
+            logger.warn("    ⚠️ LLM returned empty response. Skipping epoch step.");
+            continue;
+        }
+
+        // Clean LLM JSON response from markdown wrappers if present
+        let cleanedJson = llmResponse.trim();
+        if (cleanedJson.startsWith("```json")) {
+            cleanedJson = cleanedJson.replace(/^```json/, "").replace(/```$/, "").trim();
+        } else if (cleanedJson.startsWith("```")) {
+            cleanedJson = cleanedJson.replace(/^```/, "").replace(/```$/, "").trim();
+        }
+
+        // 3. Invoke Rust binary for optimizing step
+        logger.log(`    ${logger.c('gray', '●')} Merging and applying bounded edits via Rust Core...`);
+        let optResultStr = "";
+        try {
+            const tempFile = path.default.join(os.default.tmpdir(), `edits-${Date.now()}.json`);
+            fs.default.writeFileSync(tempFile, cleanedJson, 'utf8');
+            
+            const result = spawnSync(binPath, [
+                "optimize-step",
+                "--skill-path",
+                skillPath,
+                "--edits-json",
+                cleanedJson,
+                "--budget",
+                String(editBudget)
+            ], { encoding: 'utf8', cwd: targetDir, stdio: 'pipe' });
+            
+            if (result.error) {
+                throw result.error;
+            }
+            optResultStr = result.stdout || "";
+            
+            try { fs.default.unlinkSync(tempFile); } catch {}
+        } catch (e) {
+            logger.err(`    ✖ Rust Core optimization failed: ${e.message}`);
+            continue;
+        }
+
+        let optReport;
+        try {
+            optReport = JSON.parse(optResultStr.trim());
+        } catch {
+            logger.err("    ✖ Failed to parse optimizer report from Rust Core.");
+            continue;
+        }
+
+        if (optReport.applied_count === 0) {
+            logger.log(`    ${logger.c('yellow', '●')} No edits applied (deduplicated or target not found).`);
+            continue;
+        }
+
+        logger.log(`    ${logger.c('green', '●')} Applied ${optReport.applied_count} edits to skill.`);
+
+        // 4. Run Validation Harness
+        logger.log(`    ${logger.c('gray', '●')} Evaluating candidate skill against validation split...`);
+        let valSuccess = true;
+        try {
+            const args = harnessCmd.trim().split(/\s+/);
+            const cmd = args.shift();
+            execFileSync(cmd, args, { encoding: 'utf8', cwd: targetDir, stdio: 'pipe' });
+        } catch {
+            valSuccess = false;
+        }
+
+        const candidateScore = valSuccess ? 1.0 : 0.0;
+
+        // 5. Validation Gate
+        if (candidateScore > bestScore || (candidateScore === 1.0 && bestScore === 0.0)) {
+            logger.ok(`    ✔ Gate Accepted! Score improved/stabilized to ${logger.c('green', String(candidateScore))}.`);
+            bestScore = candidateScore;
+            currentSkillText = fs.default.readFileSync(skillPath, 'utf8');
+        } else {
+            logger.warn(`    ⚠️ Gate Rejected. Score (${candidateScore}) <= Best (${bestScore}). Reverting edits.`);
+            fs.default.writeFileSync(skillPath, currentSkillText, 'utf8');
+            
+            // Save to rejected-edit buffer history
+            const rejectedPath = path.default.join(agentDest, 'history', 'skill-optimization');
+            fs.default.mkdirSync(rejectedPath, { recursive: true });
+            const bufFile = path.default.join(rejectedPath, 'rejected-buffer.json');
+            let buffer = [];
+            if (fs.default.existsSync(bufFile)) {
+                try { buffer = JSON.parse(fs.default.readFileSync(bufFile, 'utf8')); } catch {}
+            }
+            buffer.push({ epoch, timestamp: new Date().toISOString(), edits: optReport.reports });
+            fs.default.writeFileSync(bufFile, JSON.stringify(buffer, null, 2), 'utf8');
+        }
+        console.log();
+    }
+
+    logger.ok(`🎉 SkillOpt training complete. Best validation score: ${logger.c('green', String(bestScore))}`);
+    console.log();
+}