closed-loop-cli 1.0.3 → 1.0.4

package/dist/orchestrator/prompt-optimizer.js

@@ -1,169 +0,0 @@
-"use strict";
-var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
-    if (k2 === undefined) k2 = k;
-    var desc = Object.getOwnPropertyDescriptor(m, k);
-    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
-      desc = { enumerable: true, get: function() { return m[k]; } };
-    }
-    Object.defineProperty(o, k2, desc);
-}) : (function(o, m, k, k2) {
-    if (k2 === undefined) k2 = k;
-    o[k2] = m[k];
-}));
-var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
-    Object.defineProperty(o, "default", { enumerable: true, value: v });
-}) : function(o, v) {
-    o["default"] = v;
-});
-var __importStar = (this && this.__importStar) || (function () {
-    var ownKeys = function(o) {
-        ownKeys = Object.getOwnPropertyNames || function (o) {
-            var ar = [];
-            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
-            return ar;
-        };
-        return ownKeys(o);
-    };
-    return function (mod) {
-        if (mod && mod.__esModule) return mod;
-        var result = {};
-        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
-        __setModuleDefault(result, mod);
-        return result;
-    };
-})();
-var __importDefault = (this && this.__importDefault) || function (mod) {
-    return (mod && mod.__esModule) ? mod : { "default": mod };
-};
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.runPromptOptimization = runPromptOptimization;
-const fs = __importStar(require("fs"));
-const path = __importStar(require("path"));
-const sdk_1 = __importDefault(require("@anthropic-ai/sdk"));
-const dotenv = __importStar(require("dotenv"));
-const shell_tools_1 = require("../tools/shell-tools");
-const tui_tools_1 = require("../tools/tui-tools");
-const state_manager_1 = require("./state-manager");
-const prompt_benchmark_1 = require("./prompt-benchmark");
-dotenv.config();
-const apiKey = process.env.ANTHROPIC_API_KEY || process.env.ANTHROPIC_AUTH_TOKEN || '';
-const baseURL = process.env.ANTHROPIC_BASE_URL || undefined;
-const defaultModel = process.env.ANTHROPIC_MODEL || 'mimo-v2.5-pro[1m]';
-const FALLBACK_PROMPT = `You are an elite coding agent designed to operate inside a codebase workspace.
-Your primary objective is to fulfill the user's coding tasks accurately and cleanly.
-
-You have access to a set of local tools to read, write, edit files and execute terminal commands.
-Always prefer editing precise parts of files using editFile instead of overwriting the whole file with writeFile unless it is a new file.
-When running commands, verify compilation and test outcomes. If a test fails, you must attempt to fix the issues (Self-Healing).
-
-Be concise and professional. Formulate plans before making changes.`;
-/**
- * Loads the current baseline system prompt from disk, falling back to a hardcoded default.
- */
-function loadBaselinePrompt(promptPath) {
-    if (fs.existsSync(promptPath)) {
-        return fs.readFileSync(promptPath, 'utf-8');
-    }
-    return FALLBACK_PROMPT;
-}
-/**
- * Generates an optimized candidate system prompt using an LLM refinement call.
- */
-async function generateCandidatePrompt(baselinePrompt) {
-    const anthropic = new sdk_1.default({ apiKey, baseURL });
-    const optimizerPrompt = `You are an expert system prompt engineer. Your goal is to optimize the following system prompt for an autonomous coding agent.
-You need to make the prompt more concise, remove redundancy, and explicitly instruct the agent to reduce token waste and thinking overhead, while ensuring it retains all functionality (using file-tools, rulez, self-healing, compiling, running tests).
-
-Here is the Current System Prompt:
-"""
-${baselinePrompt}
-"""
-
-Provide your output strictly as the new optimized system prompt. Do not write any markdown wrappers (like \`\`\` or \`\`\`txt), explanations, greetings, or conversational headers/footers. Output ONLY the raw optimized system prompt text.`;
-    const response = await anthropic.messages.create({
-        model: defaultModel,
-        max_tokens: 1500,
-        messages: [{ role: 'user', content: optimizerPrompt }]
-    });
-    const block = response.content.find(b => b.type === 'text');
-    if (!block || !block.text) {
-        throw new Error('Failed to retrieve system prompt candidate from LLM.');
-    }
-    return block.text.trim();
-}
-/**
- * Runs the full prompt optimization and benchmarking pipeline:
- *  1. Benchmark the baseline prompt
- *  2. Generate an optimized candidate via LLM
- *  3. Benchmark the candidate
- *  4. Compare scores and commit if improved
- */
-async function runPromptOptimization() {
-    const workspaceRoot = process.cwd();
-    const promptPath = path.join(workspaceRoot, 'src/orchestrator/system-prompt.txt');
-    const baselinePrompt = loadBaselinePrompt(promptPath);
-    // --- Step 1: Benchmark baseline ---
-    console.log(`\n\x1b[35m=== STEP 1: BENCHMARKING BASELINE SYSTEM PROMPT ===\x1b[0m`);
-    const baselineReport = await (0, prompt_benchmark_1.benchmarkPrompt)(baselinePrompt);
-    console.log(`Baseline Prompt Benchmark Result:`);
-    console.log(`- Success: ${baselineReport.success ? '\x1b[32mPASS\x1b[0m' : '\x1b[31mFAIL\x1b[0m'}`);
-    console.log(`- Time:    ${baselineReport.time.toFixed(1)}s`);
-    console.log(`- Tokens:  ${baselineReport.tokens}`);
-    console.log(`- Score:   ${baselineReport.score.toFixed(2)}`);
-    // --- Step 2: Generate candidate ---
-    console.log(`\n\x1b[35m=== STEP 2: GENERATING OPTIMIZED CANDIDATE SYSTEM PROMPT ===\x1b[0m`);
-    const optimizerSpinner = new tui_tools_1.Spinner('AGP - Optimizer: Refinement model generating optimized system prompt...');
-    optimizerSpinner.start();
-    let candidatePrompt = '';
-    try {
-        candidatePrompt = await generateCandidatePrompt(baselinePrompt);
-        optimizerSpinner.stop(true, 'AGP - Optimizer: Prompt candidate generated.');
-    }
-    catch (err) {
-        const message = err instanceof Error ? err.message : String(err);
-        optimizerSpinner.stop(false, 'AGP - Optimizer: Prompt candidate generation failed.');
-        throw new Error(`Prompt Optimization Failed: ${message}`);
-    }
-    console.log('\n\x1b[33m┌── Candidate System Prompt ──────────────────────────────────────────────────────┐\x1b[0m');
-    candidatePrompt.split('\n').forEach(line => console.log(`\x1b[33m│\x1b[0m ${line}`));
-    console.log('\x1b[33m└─────────────────────────────────────────────────────────────────────────────────┘\x1b[0m\n');
-    // --- Step 3: Benchmark candidate ---
-    console.log(`\x1b[35m=== STEP 3: BENCHMARKING CANDIDATE SYSTEM PROMPT ===\x1b[0m`);
-    const candidateReport = await (0, prompt_benchmark_1.benchmarkPrompt)(candidatePrompt);
-    console.log(`Candidate Prompt Benchmark Result:`);
-    console.log(`- Success: ${candidateReport.success ? '\x1b[32mPASS\x1b[0m' : '\x1b[31mFAIL\x1b[0m'}`);
-    console.log(`- Time:    ${candidateReport.time.toFixed(1)}s`);
-    console.log(`- Tokens:  ${candidateReport.tokens}`);
-    console.log(`- Score:   ${candidateReport.score.toFixed(2)}`);
-    // --- Step 4: Compare and commit ---
-    console.log(`\x1b[35m=== STEP 4: COMPARING PERFORMANCE METRICS ===\x1b[0m`);
-    console.log(`Baseline Score:  \x1b[1m${baselineReport.score.toFixed(2)}\x1b[0m`);
-    console.log(`Candidate Score: \x1b[1m${candidateReport.score.toFixed(2)}\x1b[0m`);
-    const approved = candidateReport.success && candidateReport.score > baselineReport.score;
-    if (approved) {
-        console.log(`\n\x1b[32;1m[Optimization Approved] Candidate prompt is more efficient. Overwriting system-prompt.txt...\x1b[0m`);
-        fs.writeFileSync(promptPath, candidatePrompt, 'utf-8');
-        // Commit the changes to Git
-        await (0, shell_tools_1.runCommand)('git add src/orchestrator/system-prompt.txt');
-        await (0, shell_tools_1.runCommand)('git commit -m "chore: optimize Task Agent system prompt via automated benchmarking loop"');
-        console.log(`\x1b[32m[Git Status] System prompt snapshot committed to repository HEAD.\x1b[0m`);
-    }
-    else {
-        console.log(`\n\x1b[31;1m[Optimization Rejected] Candidate prompt did not improve performance. Retaining baseline.\x1b[0m`);
-    }
-    (0, state_manager_1.updateEvolutionState)({
-        lastPromptOptimization: {
-            success: candidateReport.success,
-            baselineScore: baselineReport.score,
-            candidateScore: candidateReport.score,
-            approved,
-            time: new Date().toISOString()
-        }
-    });
-    (0, state_manager_1.appendHistory)({
-        cycle: (0, state_manager_1.getEvolutionState)().currentCycle,
-        type: 'prompt_opt',
-        description: `Optimized system prompt. Baseline: ${baselineReport.score.toFixed(2)}, Candidate: ${candidateReport.score.toFixed(2)}`,
-        success: approved
-    });
-}

package/dist/orchestrator/refactor-scanner.js

@@ -1,222 +0,0 @@
-"use strict";
-var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
-    if (k2 === undefined) k2 = k;
-    var desc = Object.getOwnPropertyDescriptor(m, k);
-    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
-      desc = { enumerable: true, get: function() { return m[k]; } };
-    }
-    Object.defineProperty(o, k2, desc);
-}) : (function(o, m, k, k2) {
-    if (k2 === undefined) k2 = k;
-    o[k2] = m[k];
-}));
-var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
-    Object.defineProperty(o, "default", { enumerable: true, value: v });
-}) : function(o, v) {
-    o["default"] = v;
-});
-var __importStar = (this && this.__importStar) || (function () {
-    var ownKeys = function(o) {
-        ownKeys = Object.getOwnPropertyNames || function (o) {
-            var ar = [];
-            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
-            return ar;
-        };
-        return ownKeys(o);
-    };
-    return function (mod) {
-        if (mod && mod.__esModule) return mod;
-        var result = {};
-        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
-        __setModuleDefault(result, mod);
-        return result;
-    };
-})();
-var __importDefault = (this && this.__importDefault) || function (mod) {
-    return (mod && mod.__esModule) ? mod : { "default": mod };
-};
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.getAutonomousRefactorProposal = getAutonomousRefactorProposal;
-const fs = __importStar(require("fs"));
-const path = __importStar(require("path"));
-const sdk_1 = __importDefault(require("@anthropic-ai/sdk"));
-const dotenv = __importStar(require("dotenv"));
-const tui_tools_1 = require("../tools/tui-tools");
-const repo_map_1 = require("../tools/repo-map");
-dotenv.config();
-const apiKey = process.env.ANTHROPIC_API_KEY || process.env.ANTHROPIC_AUTH_TOKEN || '';
-const baseURL = process.env.ANTHROPIC_BASE_URL || undefined;
-const defaultModel = process.env.ANTHROPIC_MODEL || 'mimo-v2.5-pro[1m]';
-/**
- * Reads the last N lines of the evolution log file.
- */
-function readRecentEvolutionLog(workspaceRoot, maxLines) {
-    const logPath = path.join(workspaceRoot, 'evolution.log');
-    if (!fs.existsSync(logPath))
-        return '';
-    try {
-        const fullLog = fs.readFileSync(logPath, 'utf-8');
-        const lines = fullLog.split('\n');
-        return lines.slice(-maxLines).join('\n');
-    }
-    catch (e) {
-        return '';
-    }
-}
-/**
- * Reads the Learnings.md file if present.
- */
-function readLearnings(workspaceRoot) {
-    const learningsPath = path.join(workspaceRoot, 'Learnings.md');
-    if (!fs.existsSync(learningsPath))
-        return '';
-    try {
-        return fs.readFileSync(learningsPath, 'utf-8');
-    }
-    catch (e) {
-        return '';
-    }
-}
-/**
- * Step 1: Ask the LLM to select 2 candidate files for refactoring
- * based on the repo map, logs, and learnings.
- */
-async function selectCandidateFiles(anthropic, repoMap, evolutionLog, learningsMd) {
-    const candidatePrompt = `You are an expert software architect. Your goal is to select the top 2 files in this codebase that would benefit most from refactoring (improving type-safety, code duplication, helper functions cleanup, or performance).
-Review the codebase map, past evolution logs, and lessons learned.
-
-Codebase Repository Map:
-${repoMap}
-
-Last 40 lines of evolution logs (shows recent runs/warnings/failures):
-${evolutionLog || 'No log history available.'}
-
-Learnings Registry (shows past insights):
-${learningsMd || 'No past learnings recorded.'}
-
-Output your response strictly as a JSON object containing a list of 2 candidate target files (relative paths). Do not include any other text, markdown blocks, or formatting, just the raw JSON.
-
-Format:
-{
-  "candidates": [
-    { "targetFile": "relative/path/to/file1.ts" },
-    { "targetFile": "relative/path/to/file2.ts" }
-  ]
-}`;
-    const response = await anthropic.messages.create({
-        model: defaultModel,
-        max_tokens: 4000,
-        messages: [{ role: 'user', content: candidatePrompt }]
-    });
-    const block = response.content.find(b => b.type === 'text');
-    if (!block || !block.text) {
-        console.error('DEBUG - candidateResponse content:', JSON.stringify(response.content));
-        throw new Error('Failed to retrieve candidates from LLM.');
-    }
-    const text = block.text.trim().replace(/^```json/, '').replace(/```$/, '').trim();
-    const json = JSON.parse(text);
-    const candidates = json.candidates || [];
-    if (candidates.length === 0) {
-        throw new Error('No candidates identified by the LLM.');
-    }
-    return candidates;
-}
-/**
- * Reads the contents of candidate files from disk for context.
- */
-function readCandidateFileContents(workspaceRoot, candidates) {
-    let fileContext = '';
-    for (const cand of candidates) {
-        const fullPath = path.join(workspaceRoot, cand.targetFile);
-        if (fs.existsSync(fullPath)) {
-            try {
-                const fileContent = fs.readFileSync(fullPath, 'utf-8');
-                fileContext += `\n--- FILE: ${cand.targetFile} ---\n${fileContent}\n`;
-            }
-            catch (e) {
-                fileContext += `\n--- FILE: ${cand.targetFile} (Could not read file contents) ---\n`;
-            }
-        }
-        else {
-            fileContext += `\n--- FILE: ${cand.targetFile} (File not found on disk) ---\n`;
-        }
-    }
-    return fileContext;
-}
-/**
- * Step 2: Ask the LLM to evaluate the candidate file contents and select the single
- * best refactoring proposal.
- */
-async function evaluateAndSelectWinner(anthropic, fileContext) {
-    const evaluationPrompt = `You are an expert software architect. Analyze the actual code content of the candidate files below and select the single best refactoring proposal.
-Evaluate each file and design a specific refactor goal.
-Score each proposal out of 10 based on:
-1. Impact (readability, duplication reduction, safety, performance).
-2. Feasibility (compilation safety, low risk of breaking tests).
-
-Candidate File Contents:
-${fileContext}
-
-Output your response strictly as a JSON object containing the winning proposal's target file and the specific refactoring goal. Do not include any other text, markdown blocks, or formatting, just the raw JSON.
-
-Format:
-{
-  "winningProposal": {
-    "targetFile": "relative/path/to/file.ts",
-    "refactorGoal": "A concise, specific description of the refactoring goal (e.g., 'Refactor search-tools to use strong interfaces instead of any[]')",
-    "impactScore": 8.5,
-    "feasibilityScore": 9.0
-  }
-}`;
-    const response = await anthropic.messages.create({
-        model: defaultModel,
-        max_tokens: 4000,
-        messages: [{ role: 'user', content: evaluationPrompt }]
-    });
-    const block = response.content.find(b => b.type === 'text');
-    if (!block || !block.text) {
-        throw new Error('Failed to retrieve evaluation from LLM.');
-    }
-    const text = block.text.trim().replace(/^```json/, '').replace(/```$/, '').trim();
-    const json = JSON.parse(text);
-    const winningProposal = json.winningProposal;
-    if (!winningProposal || !winningProposal.targetFile || !winningProposal.refactorGoal) {
-        throw new Error('Invalid JSON structure returned for winning proposal.');
-    }
-    return {
-        targetFile: winningProposal.targetFile,
-        refactorGoal: winningProposal.refactorGoal
-    };
-}
-/**
- * Generates a refactoring proposal by scanning the codebase repository map,
- * evolution logs, and learnings. Uses a two-pass LLM strategy:
- *  1. Select candidate files
- *  2. Evaluate actual code and pick the winner
- */
-async function getAutonomousRefactorProposal() {
-    const workspaceRoot = process.cwd();
-    const repoMap = (0, repo_map_1.generateRepoMap)(workspaceRoot);
-    const scanSpinner = new tui_tools_1.Spinner('AGP - Scan: Auditing codebase structure for optimizations...');
-    scanSpinner.start();
-    // Context gathering
-    const evolutionLog = readRecentEvolutionLog(workspaceRoot, 40);
-    const learningsMd = readLearnings(workspaceRoot);
-    const anthropic = new sdk_1.default({ apiKey, baseURL });
-    try {
-        // Step 1: Select 2 target candidate files
-        const candidates = await selectCandidateFiles(anthropic, repoMap, evolutionLog, learningsMd);
-        // Step 2: Read candidate file contents
-        const fileContext = readCandidateFileContents(workspaceRoot, candidates);
-        // Step 3: Evaluate and select winner
-        const proposal = await evaluateAndSelectWinner(anthropic, fileContext);
-        console.log(`\n\x1b[32m✔\x1b[0m AGP - Scan: Selected best proposal: ${proposal.targetFile}`);
-        scanSpinner.stop(true, 'AGP - Scan: Codebase audit complete.');
-        return proposal;
-    }
-    catch (err) {
-        const message = err instanceof Error ? err.message : String(err);
-        scanSpinner.stop(false, 'AGP - Scan: Codebase audit failed.');
-        throw new Error(`Refactoring Analysis Failed: ${message}`);
-    }
-}