@mndrk/agx 1.4.62 → 1.4.63

package/lib/cli/cloud/prompts.js

@@ -1,6 +1,8 @@
 /* eslint-disable no-console */
 'use strict';
 
+const { interpolate, AGGREGATOR_PROMPT, EXECUTE_ITERATION, VERIFY_PROMPT } = require('../../prompts/templates');
+
 function createCloudPromptHelpers(env) {
   const { path, truncateForPrompt, VERIFY_PROMPT_MAX_CHARS } = env || {};
 
@@ -49,57 +51,20 @@ function createCloudPromptHelpers(env) {
         '- artifacts/ (additional artifacts, if any)',
       ].join('\n');
 
-    return `You are the decision aggregator for a ${role} run.
-
-Task ID: ${taskId}
-Title: ${task?.title || taskId}
-Stage: ${task?.stage || 'unknown'}
-
-User Request: 
-"""
-${task?.title}
-${task?.content}
----
-Task Thread:
-${taskComments.map(c => `${c.author}: ${c.content}`).join('\n')}
-"""
-
-Stage Objective: ${stagePrompt}
-Stage Completion Requirement: ${stageRequirement}
-
-Local run artifacts folder: ${runRoot || '(not available)'}
-Key run files:
-${runFiles}
-
-Relevant files referenced during execution (detected from output/logs):
-${refsBlock}
-
-Decide if the task is done. If not, provide the next instruction for another iteration.
-Only set "done": true when the Stage Completion Requirement is satisfied.
-
-You may think through your analysis first, but you MUST end your response with valid JSON.
-
-Output contract (strict):
-- You may include thinking/reasoning at the start of your response
-- Your response MUST end with exactly one raw JSON object
-- Do not use markdown/code fences/backticks around the JSON
-- Do not add commentary after the JSON
-- Use double-quoted keys and strings
-- Keep newlines escaped inside strings
-- If "done" is false, "next_prompt" must be a non-empty actionable instruction
-
-The final JSON in your response must have this exact shape:
-{
-  "done": false,
-  "decision": "done|blocked|not_done|failed",
-  "explanation": "clear explanation of the decision",
-  "final_result": "final result if done, empty string otherwise",
-  "next_prompt": "specific actionable instruction for next iteration",
-  "summary": "brief summary of current state"
-}
-
-If uncertain, still return valid JSON with decision "failed" and explain why in "explanation".
-`;
+    return interpolate(AGGREGATOR_PROMPT, {
+      role,
+      taskId,
+      title: task?.title || taskId,
+      stage: task?.stage || 'unknown',
+      taskTitle: task?.title,
+      taskContent: task?.content,
+      taskComments: taskComments.map(c => `${c.author}: ${c.content}`).join('\n'),
+      stagePrompt,
+      stageRequirement,
+      runRoot: runRoot || '(not available)',
+      runFiles,
+      refsBlock,
+    });
   }
 
   function truncateForPromptLocal(text, maxChars) {
@@ -114,26 +79,10 @@ If uncertain, still return valid JSON with decision "failed" and explain why in
       ? nextPrompt.trim()
       : 'Pick the next concrete step and implement it.';
 
-    return [
-      'EXECUTE PHASE',
-      `Iteration: ${iteration}`,
-      '',
-      'Keep output concise and avoid dumping full file contents or long logs.',
-      'If you need to reference code, cite paths and describe changes instead of pasting whole files.',
-      '',
-      'Output contract:',
-      '- Start with "PLAN:" then 2-5 bullets.',
-      '- Do the work.',
-      '- End with "IMPLEMENTATION SUMMARY:" bullets:',
-      '  - Changed: (paths only, 10 max)',
-      '  - Commands: (what you ran)',
-      '  - Notes:',
-      '',
-      `Task for this iteration: ${instruction}`,
-      '',
-      'Do not output JSON in this phase.',
-      ''
-    ].join('\n');
+    return interpolate(EXECUTE_ITERATION, {
+      iteration,
+      instruction,
+    });
   }
 
   function buildVerifyPrompt({ taskId, task, stagePrompt, stageRequirement, gitSummary, verifyResults, iteration, lastRunPath, agentOutput }) {
@@ -161,68 +110,21 @@ If uncertain, still return valid JSON with decision "failed" and explain why in
 
     const runRoot = lastRunPath ? String(lastRunPath) : '';
 
-    const prompt = `You are the verifier for an agx run.
-
-Task ID: ${taskId}
-Title: ${title || taskId}
-Stage: ${task?.stage || 'unknown'}
-Iteration: ${iteration}
-
-Stage Objective: ${stagePrompt}
-Stage Completion Requirement: ${stageRequirement}
-
-Local run artifacts folder: ${runRoot || '(not available)'}
-
-User Request:
-"""
-${title}
-${content}
-"""
-
-Target state evidence:
-- Before reviewing git status or diff, first ask whether the target state already exists and fully satisfies the user request.
-- Check the agent output and verification command logs for evidence (files, folders, config values, summaries) that the desired outcome is already in place, even if the diff is empty.
-- If those signals show the target is already achieved and verification commands pass, treat the work as done without requiring new diffs; only rely on git status/diff when the target state is missing or unclear.
-
-Repo summary (git):
-Status (porcelain):
-${statusShort || '(none)'}
-
-Diff (stat):
-${diffShort || '(none)'}
-
-Verification commands:
-${cmdLines}
-
-Agent output (last iteration):
-${agentOutputShort || '(not available)'}
-
-Decide if the stage is complete. Use verification commands as evidence.
-Ignore unrelated working tree changes; focus on whether the user request is satisfied.
-If not complete, provide the next smallest instruction for another iteration.
-Set "done": true when the user request is satisfied and the evidence supports it. Treat the stage objective/requirement as guidance, not a keyword checklist.
-
-IMPORTANT: Before you look at the git status/diff, first ask whether the target state already exists and the user request is satisfied. Check the agent output and verification command logs for evidence (files, folders, config values, summaries) that the outcome is already in place, even if the diff is empty. Treat that as "target achieved" and you may mark the stage done without requiring new diffs. Use git status/diff only when the target state is missing or unclear and you need to verify what changed to reach it. If verification commands pass and the evidence shows the desired result already exists, mark the work done.
-
-Output contract (strict): your response MUST be exactly one raw JSON object with this shape:
-{
-  "done": false,
-  "decision": "done|blocked|not_done|failed",
-  "explanation": "clear explanation of the decision",
-  "final_result": "final result if done, empty string otherwise",
-  "next_prompt": "specific actionable instruction for next iteration",
-  "summary": "brief summary of current state",
-  "plan_md": "PLAN markdown for this iteration (newlines escaped)",
-  "implementation_summary_md": "IMPLEMENTATION SUMMARY markdown (newlines escaped)",
-  "verification_md": "VERIFICATION markdown (newlines escaped)"
-}
-
-Rules:
-- Use double-quoted keys and strings.
-- Keep newlines escaped inside strings (use \\n).
-- Keep the markdown fields short and checklist-style.
-- Always fill "explanation". For "blocked", include what is blocking and what input/action would unblock. For "failed", include what failed (command/tool/error) and a recovery step.
-`;
+    const prompt = interpolate(VERIFY_PROMPT, {
+      taskId,
+      title: title || taskId,
+      stage: task?.stage || 'unknown',
+      iteration,
+      stagePrompt,
+      stageRequirement,
+      runRoot: runRoot || '(not available)',
+      requestTitle: title,
+      requestContent: content,
+      statusShort: statusShort || '(none)',
+      diffShort: diffShort || '(none)',
+      cmdLines,
+      agentOutputShort: agentOutputShort || '(not available)',
+    });
 
     return (typeof truncateForPrompt === 'function'
       ? truncateForPrompt(prompt, VERIFY_PROMPT_MAX_CHARS)

package/lib/cli/runCli.js

@@ -102,6 +102,7 @@ const {
 const { createCloudRunner } = require('./cloud');
 const daemon = require('./daemon');
 const { loadTemplate, renderTemplateContent } = require('../templates');
+const { applyTemplateDefaults } = require('./templateDefaults');
 const { slugify } = require('../storage/paths');
 const {
   DAEMON_PID_FILE,
@@ -1343,20 +1344,6 @@ async function checkOnboarding() {
       templateName = args[templateIdx + 1];
     }
 
-    // Default provider from config
-    if (!provider) {
-      const config = loadConfig();
-      provider = config?.defaultProvider || 'claude';
-    }
-    if (template) {
-      if (!providerArgProvided && template.provider) {
-        provider = template.provider;
-      }
-      if (!modelArgProvided && template.model) {
-        model = template.model;
-      }
-    }
-
     // Extract goal text (filter out flags)
     const flagsToRemove = [
       '--json',
@@ -1412,6 +1399,18 @@ async function checkOnboarding() {
       }
     }
 
+    const config = loadConfig();
+    const defaults = applyTemplateDefaults({
+      provider,
+      model,
+      template,
+      providerArgProvided,
+      modelArgProvided,
+      defaultProvider: config?.defaultProvider || 'claude',
+    });
+    provider = defaults.provider;
+    model = defaults.model;
+
     try {
       let projectId = null;
       if (projectSlug) {
@@ -1871,7 +1870,7 @@ async function checkOnboarding() {
     }
   }
 
-  // agx task ls [-a]  (Docker-style namespace)
+  // agx task ls [-a] [search] (Docker-style namespace)
   if ((cmd === 'task' && args[1] === 'ls') || cmd === 'list' || cmd === 'ls' || cmd === 'tasks') {
     const showAll = args.includes('-a') || args.includes('--all');
     const config = loadCloudConfig();
@@ -1880,10 +1879,42 @@ async function checkOnboarding() {
       process.exit(1);
     }
 
+    // Extract search term
+    const flagsToSkip = ['-a', '--all'];
+    const cmdTokens = (cmd === 'task' && args[1] === 'ls') ? ['task', 'ls'] : [cmd];
+    const searchParts = args.filter((arg, i) => {
+      // Skip flags
+      if (arg.startsWith('-')) return false;
+      // Skip command words (simple heuristic: first occurrence of cmd words)
+      // Actually simpler: iterate args, skip known command words at start.
+      return true;
+    });
+
+    // Better extraction:
+    let startIndex = (cmd === 'task' && args[1] === 'ls') ? 2 : 1;
+    let search = '';
+    const searchTerms = [];
+    for (let i = startIndex; i < args.length; i++) {
+      if (!args[i].startsWith('-')) {
+        searchTerms.push(args[i]);
+      }
+    }
+    if (searchTerms.length > 0) {
+      search = searchTerms.join(' ');
+    }
+
     try {
-      const { tasks } = await cloudRequest('GET', '/api/tasks');
+      let url = '/api/tasks';
+      if (search) {
+        url += `?search=${encodeURIComponent(search)}`;
+      }
+      const { tasks } = await cloudRequest('GET', url);
       if (tasks.length === 0) {
-        console.log(`${c.dim}No tasks in queue${c.reset}`);
+        if (search) {
+           console.log(`${c.dim}No tasks found matching "${search}"${c.reset}`);
+        } else {
+           console.log(`${c.dim}No tasks in queue${c.reset}`);
+        }
         process.exit(0);
       }
 
@@ -2839,6 +2870,117 @@ async function checkOnboarding() {
     process.exit(0);
   }
 
+  // ============================================================
+  // agx feedback - Quick bug reports with context
+  // ============================================================
+  if (cmd === 'feedback') {
+    const readline = require('readline');
+    const { exec, execSync } = require('child_process');
+    
+    // Get description from args or prompt
+    let description = args.slice(1).join(' ');
+    if (!description) {
+      const rl = readline.createInterface({ input: process.stdin, output: process.stdout });
+      description = await new Promise(resolve => {
+        rl.question(`${c.cyan}What went wrong?${c.reset} `, answer => {
+          rl.close();
+          resolve(answer.trim());
+        });
+      });
+    }
+    
+    if (!description) {
+      console.log(`${c.yellow}No description provided. Aborting.${c.reset}`);
+      process.exit(1);
+    }
+    
+    console.log(`\n${c.dim}Collecting context...${c.reset}`);
+    
+    // Collect system info
+    const version = (() => { try { return require('../../package.json').version; } catch { return 'unknown'; } })();
+    const nodeVersion = process.version;
+    const platform = `${os.platform()} ${os.arch()}`;
+    
+    // Collect recent daemon logs (sanitized)
+    let recentLogs = '';
+    const logDir = path.join(os.homedir(), '.agx', 'logs');
+    try {
+      const logFiles = fs.readdirSync(logDir).filter(f => f.endsWith('.log')).sort().reverse();
+      if (logFiles.length > 0) {
+        const latestLog = fs.readFileSync(path.join(logDir, logFiles[0]), 'utf8');
+        const lines = latestLog.split('\n').slice(-30).join('\n');
+        // Sanitize: remove potential secrets, paths with usernames
+        recentLogs = lines
+          .replace(/\/Users\/[^\/\s]+/g, '/Users/***')
+          .replace(/\/home\/[^\/\s]+/g, '/home/***')
+          .replace(/sk-[a-zA-Z0-9]+/g, 'sk-***')
+          .replace(/key["\s:=]+[a-zA-Z0-9]+/gi, 'key=***');
+      }
+    } catch { }
+    
+    // Build issue body
+    const issueTitle = description.length > 60 ? description.substring(0, 60) + '...' : description;
+    const issueBody = `## Description
+${description}
+
+## Environment
+- **agx version:** ${version}
+- **Node.js:** ${nodeVersion}
+- **Platform:** ${platform}
+
+## Recent Logs
+\`\`\`
+${recentLogs || 'No logs available'}
+\`\`\`
+
+---
+*Submitted via \`agx feedback\`*`;
+
+    // Try gh CLI first
+    let ghAvailable = false;
+    try {
+      execSync('gh --version', { stdio: 'ignore' });
+      ghAvailable = true;
+    } catch { }
+    
+    const repoUrl = 'https://github.com/ramarlina/agx';
+    
+    if (ghAvailable) {
+      console.log(`${c.dim}Using GitHub CLI...${c.reset}`);
+      try {
+        const result = execSync(
+          `gh issue create --repo ramarlina/agx --title "${issueTitle.replace(/"/g, '\\"')}" --body "${issueBody.replace(/"/g, '\\"').replace(/\n/g, '\\n')}"`,
+          { encoding: 'utf8', stdio: ['pipe', 'pipe', 'pipe'] }
+        );
+        console.log(`\n${c.green}✓${c.reset} Issue created: ${result.trim()}`);
+      } catch (err) {
+        console.log(`${c.yellow}GitHub CLI failed, opening browser instead...${c.reset}`);
+        ghAvailable = false;
+      }
+    }
+    
+    if (!ghAvailable) {
+      // Fall back to opening browser with pre-filled URL
+      const params = new URLSearchParams({
+        title: issueTitle,
+        body: issueBody,
+        labels: 'bug,user-feedback'
+      });
+      const issueUrl = `${repoUrl}/issues/new?${params.toString()}`;
+      
+      const openCmd = os.platform() === 'darwin' ? 'open' : os.platform() === 'win32' ? 'start' : 'xdg-open';
+      try {
+        execSync(`${openCmd} "${issueUrl}"`, { stdio: 'ignore' });
+        console.log(`\n${c.green}✓${c.reset} Opened GitHub issue form in browser`);
+      } catch {
+        console.log(`\n${c.cyan}Open this URL to submit:${c.reset}\n${issueUrl}`);
+      }
+    }
+    
+    console.log(`\n${c.dim}Thanks for the feedback! 🙏${c.reset}`);
+    process.exit(0);
+  }
+
   // ============================================================
   // agx telemetry - Manage anonymous usage telemetry
   // ============================================================

package/lib/cli/skillText.js

@@ -1,60 +1,5 @@
 // agx skill - instructions for LLMs on how to use agx
 
-const AGX_SKILL = `---
-name: agx
-description: Task orchestrator for AI agents. Uses cloud API for persistence.
----
-
-# agx - AI Agent Task Orchestrator
-
-agx manages tasks and coordinates AI agents. Uses cloud API for persistence.
-
-## Quick Start
-
-\`\`\`bash
-agx -a -p "Build a REST API"  # Autonomous: works until done
-agx -p "explain this code"     # One-shot question
-\`\`\`
-
-## Task Lifecycle
-
-\`\`\`bash
-agx new "goal"          # Create task
-agx run [task]          # Run a task
-agx complete <taskId>   # Mark task stage complete
-agx status <taskId>     # Show detailed info for a specific task
-agx status              # Show current status
-\`\`\`
-
-## Checking Tasks
-
-\`\`\`bash
-agx task ls             # List tasks
-agx task logs <id> [-f] # View/tail task logs
-agx task tail <id>      # Tail task logs
-agx comments tail <id>  # Tail task comments
-agx logs tail <id>      # Tail task logs
-agx watch               # Watch task updates in real-time (SSE)
-\`\`\`
-
-## Cloud
-
-\`\`\`bash
-AGX_CLOUD_URL=http://localhost:41741 agx status
-AGX_CLOUD_URL=http://localhost:41741 agx task ls
-agx daemon start  # Start local daemon
-\`\`\`
-
-## Providers
-
-claude (c), gemini (g), ollama (o), codex (x)
-
-## Key Flags
-
--a  Autonomous mode (daemon + work until done)
--p  Prompt/goal
--y  Skip confirmations (implied by -a)
--P, --provider <c|g|o|x>  Provider for new task (claude/gemini/ollama/codex)
-`;
+const { AGX_SKILL } = require('../prompts/templates');
 
 module.exports = { AGX_SKILL };

package/lib/cli/templateDefaults.js

@@ -0,0 +1,22 @@
+'use strict';
+
+function applyTemplateDefaults({
+    provider,
+    model,
+    template,
+    providerArgProvided,
+    modelArgProvided,
+    defaultProvider,
+}) {
+    let resolvedProvider = provider;
+    if (!providerArgProvided) {
+        resolvedProvider = template?.provider || defaultProvider || 'claude';
+    }
+    let resolvedModel = model;
+    if (!modelArgProvided) {
+        resolvedModel = template?.model || model;
+    }
+    return { provider: resolvedProvider, model: resolvedModel };
+}
+
+module.exports = { applyTemplateDefaults };

package/lib/executor.js

@@ -10,22 +10,7 @@ const fs = require('fs');
 const path = require('path');
 const os = require('os');
 const { commandExists } = require('./proc/commandExists');
-
-const DEFAULT_STAGE_PROMPT = 'Execute this stage using the latest stage prompt from the cloud task context.';
-
-const PROJECT_CONTEXT = [
-  'AGX orchestrates autonomous AI agents with tasks defined via `agx new` and managed through `agx run`, `agx tasks`, `agx status`, and `agx context` so agents can wake, work, and sleep across sessions.',
-  'Task state lives in the cloud API (goal, criteria, progress, learnings) along with the orchestration worker (pg-boss); `agx info`/`agx context` expose structured project metadata.',
-  'Quick-start workflow: `agx new "<goal>"`, optionally `agx -a -p "<goal>"` for autonomous execution, then use `agx run`, `agx tasks`, `agx status`, and `agx context` to manage work.',
-  'Task management commands include `agx task ls`, `agx task logs`, `agx task stop`, `agx task rm`, `agx complete`, `agx pull`.',
-  'Daemon mode runs via `agx daemon start/stop/status/logs` so agents can poll for work continuously.',
-  'Providers: `agx claude` (alias `c`), `agx gemini` (alias `g`), and `agx ollama` (alias `o`).',
-  'Key flags: `-a/--autonomous`, `-p/--prompt`, `--prompt-file`, `-y/--yolo`, `--continue <task>`.',
-  'Key principles: persistent storage, criteria-driven completion, checkpoint often, ask when stuck, learn and adapt.',
-  'Agent workflow: orient on saved state, plan, execute, checkpoint, adapt to blockers, and report learnings.',
-  'State operations: define objectives via `agx new`, update learnings with `[learn: ...]`, mark completion via `agx complete`, and sync via `agx info`/`agx context`.',
-  'Project metadata: attach `--metadata key=value` and `--repo` info so `/api/projects` keeps structured context.',
-].join('\n');
+const { interpolate, DEFAULT_STAGE_PROMPT, PROJECT_CONTEXT, EXECUTOR_PROMPT } = require('./prompts/templates');
 
 // Engine configurations
 const ENGINES = {
@@ -238,30 +223,13 @@ async function executeTask(options) {
  * Build the full prompt for a stage
  */
 function buildPrompt(title, content, stage, stageConfig) {
-  return `# Task: ${title}
-
-## Stage: ${stage}
-
-${stageConfig.prompt}
-
-## Project Context
-
-${PROJECT_CONTEXT}
-
-## Task Details
-
-${content}
-
-## Instructions
-
-1. Complete the work for this stage
-2. Use [checkpoint: message] to save progress
-3. Use [learn: insight] to record learnings
-4. Use [done] when stage is complete
-5. Use [blocked: reason] if you need human help
-
-Focus on this stage only. The task will automatically advance to the next stage when complete.
-`;
+  return interpolate(EXECUTOR_PROMPT, {
+    title,
+    stage,
+    stagePrompt: stageConfig.prompt,
+    projectContext: PROJECT_CONTEXT,
+    content,
+  });
 }
 
 /**