@goodandready/dsh-cron 0.3.11 → 0.3.13

package/lib/dag.js

@@ -1,7 +1,7 @@
 /**
- * Directed Acyclic Graph (DAG) Engine for Chained Tasks (#328, #334)
+ * Directed Acyclic Graph (DAG) Engine for Chained Tasks (#328, #334, #352)
  * Provides cycle detection, topological sorting, dependency resolution,
- * and graph visualization modeling for multi-step task workflows.
+ * and graph visualization modeling for multi-step task workflows and decision nodes.
  */
 
 /**
@@ -61,9 +61,40 @@ export function parseChainTargets(target) {
   return [];
 }
 
+/**
+ * Extracts all outgoing targets from a task (onSuccess, branchRules, elseTask).
+ *
+ * @param {Object} task
+ * @returns {string[]}
+ */
+export function getTaskOutgoingTargets(task) {
+  if (!task || typeof task !== 'object') return [];
+  const targets = new Set();
+
+  for (const t of parseChainTargets(task.onSuccess)) {
+    targets.add(t);
+  }
+
+  if (Array.isArray(task.branchRules)) {
+    for (const r of task.branchRules) {
+      if (r && r.targetTask) {
+        const targetId = String(r.targetTask).trim();
+        if (targetId) targets.add(targetId);
+      }
+    }
+  }
+
+  if (task.elseTask) {
+    const elseId = String(task.elseTask).trim();
+    if (elseId) targets.add(elseId);
+  }
+
+  return Array.from(targets);
+}
+
 /**
  * Validates that adding or updating a task does not introduce cyclic dependencies.
- * Analyzes both `dependsOn` relationships and `onSuccess` chain links.
+ * Analyzes `dependsOn` relationships, `onSuccess` chain links, and `branchRules` decision nodes (#352).
  *
  * @param {Array<Object>|Map<string, Object>|Object} allTasks All current tasks
  * @param {Object} [pendingTask] Task being created or updated
@@ -98,16 +129,16 @@ export function validateDAG(allTasks, pendingTask = null) {
     if (deps.includes(id)) {
       return { ok: false, error: `Task "${task.title || id}" cannot depend on itself`, cycle: [id, id] };
     }
-    const chainTargets = parseChainTargets(task.onSuccess);
-    if (chainTargets.includes(id)) {
-      return { ok: false, error: `Task "${task.title || id}" cannot trigger itself as next step`, cycle: [id, id] };
+    const outgoingTargets = getTaskOutgoingTargets(task);
+    if (outgoingTargets.includes(id)) {
+      return { ok: false, error: `Task "${task.title || id}" cannot trigger itself as next step or branch target`, cycle: [id, id] };
     }
   }
 
   // Construct adjacency list: Edge from U to V means U must complete before V runs.
   // Sources of edges:
   // 1) dependsOn: If V depends on U, then U -> V
-  // 2) onSuccess: If U has onSuccess V, then U -> V
+  // 2) outgoing targets (onSuccess, branchRules, elseTask): If U triggers V, then U -> V
   const adj = new Map();
   for (const id of taskMap.keys()) {
     adj.set(id, new Set());
@@ -121,8 +152,8 @@ export function validateDAG(allTasks, pendingTask = null) {
       adj.get(depId).add(id);
     }
 
-    // 2) onSuccess (id -> target)
-    const targets = parseChainTargets(task.onSuccess);
+    // 2) outgoing targets (id -> target)
+    const targets = getTaskOutgoingTargets(task);
     for (const targetId of targets) {
       if (!adj.has(id)) adj.set(id, new Set());
       adj.get(id).add(targetId);
@@ -206,7 +237,7 @@ export function topologicalSort(tasks) {
         inDegree.set(id, (inDegree.get(id) || 0) + 1);
       }
     }
-    const targets = parseChainTargets(task.onSuccess);
+    const targets = getTaskOutgoingTargets(task);
     for (const targetId of targets) {
       if (taskMap.has(targetId)) {
         adj.get(id).add(targetId);
@@ -235,7 +266,7 @@ export function topologicalSort(tasks) {
 }
 
 /**
- * Builds a visual graph model (nodes and edges) for UI rendering (#334).
+ * Builds a visual graph model (nodes and edges) for UI rendering (#334, #352).
  *
  * @param {Array<Object>} tasks All relevant tasks
  * @param {Array<Object>} [snapshots] Checkpoints for the chain run
@@ -265,7 +296,7 @@ export function buildTaskGraphModel(tasks = [], snapshots = [], focusTaskId = nu
     let retryCount = 0;
 
     if (snap) {
-      nodeStatus = snap.status === 'completed' || snap.status === 'success' ? 'success' : 'failed';
+      nodeStatus = snap.status === 'completed' || snap.status === 'success' ? 'success' : (snap.status === 'skipped' ? 'skipped' : 'failed');
       durationMs = snap.durationMs || 0;
       retryCount = snap.retryCount || 0;
       if (snap.output) {
@@ -278,6 +309,8 @@ export function buildTaskGraphModel(tasks = [], snapshots = [], focusTaskId = nu
       nodeStatus = 'running';
     }
 
+    const hasBranches = Array.isArray(task.branchRules) && task.branchRules.length > 0;
+
     nodes.push({
       id,
       title: task.title || id,
@@ -288,8 +321,11 @@ export function buildTaskGraphModel(tasks = [], snapshots = [], focusTaskId = nu
       outputSnippet,
       isFocus: Boolean(focusTaskId && id === focusTaskId),
       isRetryable: nodeStatus === 'failed',
+      isDecisionNode: hasBranches,
       dependsOn: sanitizeDependsOn(task.dependsOn),
       joinPolicy: sanitizeJoinPolicy(task.joinPolicy),
+      branchRulesCount: hasBranches ? task.branchRules.length : 0,
+      elseTask: task.elseTask || null,
     });
 
     // Edges from dependsOn
@@ -303,17 +339,50 @@ export function buildTaskGraphModel(tasks = [], snapshots = [], focusTaskId = nu
       });
     }
 
-    // Edges from onSuccess
-    const targets = parseChainTargets(task.onSuccess);
-    for (const targetId of targets) {
+    // Edges from branchRules (#352 Decision Nodes)
+    if (hasBranches) {
+      for (const rule of task.branchRules) {
+        if (rule && rule.targetTask) {
+          const targetId = String(rule.targetTask).trim();
+          edges.push({
+            from: id,
+            to: targetId,
+            type: 'branch',
+            label: rule.label || `[if: ${rule.field} ${rule.operator} ${rule.value ?? ''}]`.trim(),
+            condition: {
+              field: rule.field,
+              operator: rule.operator,
+              value: rule.value,
+            },
+          });
+        }
+      }
+    }
+
+    // Edge from elseTask
+    if (task.elseTask) {
       edges.push({
         from: id,
-        to: targetId,
-        type: 'onSuccess',
-        label: 'onSuccess',
+        to: String(task.elseTask).trim(),
+        type: 'branch_else',
+        label: '[else]',
+        isElse: true,
       });
     }
 
+    // Edges from onSuccess (only if no branchRules to avoid redundant edge clutter)
+    if (!hasBranches) {
+      const targets = parseChainTargets(task.onSuccess);
+      for (const targetId of targets) {
+        edges.push({
+          from: id,
+          to: targetId,
+          type: 'onSuccess',
+          label: 'onSuccess',
+        });
+      }
+    }
+
     // Edges from onFailure
     const failTargets = parseChainTargets(task.onFailure);
     for (const failId of failTargets) {
@@ -328,3 +397,4 @@ export function buildTaskGraphModel(tasks = [], snapshots = [], focusTaskId = nu
 
   return { nodes, edges, hasFailure };
 }
+

package/lib/doctor.js

@@ -0,0 +1,431 @@
+/**
+ * Self-Healing Doctor Engine (#353)
+ *
+ * Provides autonomous failure classification, root-cause diagnosis,
+ * code/prompt fix proposals with unified diff previews, and safe 1-click & auto-apply healing.
+ */
+
+import { parseJsonAnswer } from './llm-ask.js';
+
+/**
+ * Standard failure categories for structured diagnosis.
+ */
+export const FAILURE_CATEGORIES = {
+  SYNTAX_ERROR: 'syntax_error',
+  DEPENDENCY_MISSING: 'dependency_missing',
+  RATE_LIMIT: 'rate_limit',
+  AUTH_FAILED: 'auth_failed',
+  SCHEMA_CHANGE: 'schema_change',
+  TIMEOUT: 'timeout',
+  NETWORK_ERROR: 'network_error',
+  RUNTIME_ERROR: 'runtime_error',
+  VALIDATION_FAILED: 'validation_failed',
+};
+
+/**
+ * Classifies an error and execution traces into a standard failure category using deterministic pattern matching.
+ *
+ * @param {Object} task
+ * @param {Object} runInfo
+ * @returns {{ category: string, matchedRule: string }}
+ */
+export function classifyFailure(task, runInfo = {}) {
+  const text = [
+    String(runInfo.error || ''),
+    String(runInfo.stderr || ''),
+    String(runInfo.output || ''),
+    String(runInfo.status || ''),
+  ].join(' ').toLowerCase();
+
+  const exitCode = runInfo.exitCode ?? null;
+
+  // 1. Timeout
+  if (runInfo.status === 'timeout' || text.includes('timed out') || text.includes('etimedout') || text.includes('sigterm') && exitCode === 143) {
+    return { category: FAILURE_CATEGORIES.TIMEOUT, matchedRule: 'timeout_detector' };
+  }
+
+  // 2. Auth / Permissions
+  if (text.includes('401 unauthorized') || text.includes('403 forbidden') || text.includes('permission denied') || text.includes('invalid api key') || text.includes('token expired')) {
+    return { category: FAILURE_CATEGORIES.AUTH_FAILED, matchedRule: 'auth_forbidden_detector' };
+  }
+
+  // 3. Rate Limit / Quota
+  if (text.includes('429') || text.includes('rate limit') || text.includes('too many requests') || text.includes('quota exceeded') || text.includes('tpm') || text.includes('rpm')) {
+    return { category: FAILURE_CATEGORIES.RATE_LIMIT, matchedRule: 'rate_limit_detector' };
+  }
+
+  // 4. Missing Dependencies / Commands
+  if (text.includes('command not found') || text.includes('no such file or directory') || text.includes('modulenotfounderror') || text.includes('cannot find module') || text.includes('not recognized as an internal or external command')) {
+    return { category: FAILURE_CATEGORIES.DEPENDENCY_MISSING, matchedRule: 'missing_dependency_detector' };
+  }
+
+  // 5. Syntax Error
+  if (text.includes('syntaxerror') || text.includes('indentationerror') || text.includes('unexpected token') || text.includes('syntax error near') || text.includes('parse error')) {
+    return { category: FAILURE_CATEGORIES.SYNTAX_ERROR, matchedRule: 'syntax_error_detector' };
+  }
+
+  // 6. Schema / JSON / Parsing mismatch
+  if (text.includes('json.parse') || text.includes('unexpected end of json') || text.includes('cannot read property') || text.includes('cannot read properties of undefined') || text.includes('keyerror')) {
+    return { category: FAILURE_CATEGORIES.SCHEMA_CHANGE, matchedRule: 'schema_mismatch_detector' };
+  }
+
+  // 7. Network / Connection
+  if (text.includes('econnrefused') || text.includes('enotfound') || text.includes('network is unreachable') || text.includes('connection reset')) {
+    return { category: FAILURE_CATEGORIES.NETWORK_ERROR, matchedRule: 'network_failure_detector' };
+  }
+
+  // 8. Validation Failed
+  if (runInfo.status === 'validation_failed' || text.includes('validation_failed')) {
+    return { category: FAILURE_CATEGORIES.VALIDATION_FAILED, matchedRule: 'validation_rule_detector' };
+  }
+
+  return { category: FAILURE_CATEGORIES.RUNTIME_ERROR, matchedRule: 'fallback_runtime_detector' };
+}
+
+/**
+ * Generates a clean text unified diff between original and proposed code.
+ *
+ * @param {string} originalText
+ * @param {string} newText
+ * @param {string} [filename]
+ * @returns {string}
+ */
+export function generateUnifiedDiff(originalText = '', newText = '', filename = 'task_script') {
+  const origLines = String(originalText).split('\n');
+  const newLines = String(newText).split('\n');
+
+  if (originalText === newText) {
+    return `--- a/${filename}\n+++ b/${filename}\n (no changes)`;
+  }
+
+  const diffLines = [
+    `--- a/${filename}`,
+    `+++ b/${filename}`,
+    `@@ -1,${origLines.length} +1,${newLines.length} @@`,
+  ];
+
+  // Simple line-by-line diff representation
+  let i = 0;
+  let j = 0;
+  while (i < origLines.length || j < newLines.length) {
+    if (i < origLines.length && j < newLines.length) {
+      if (origLines[i] === newLines[j]) {
+        diffLines.push(` ${origLines[i]}`);
+        i++;
+        j++;
+      } else {
+        diffLines.push(`-${origLines[i]}`);
+        diffLines.push(`+${newLines[j]}`);
+        i++;
+        j++;
+      }
+    } else if (i < origLines.length) {
+      diffLines.push(`-${origLines[i]}`);
+      i++;
+    } else if (j < newLines.length) {
+      diffLines.push(`+${newLines[j]}`);
+      j++;
+    }
+  }
+
+  return diffLines.join('\n');
+}
+
+/**
+ * Builds the diagnostic prompt for the model.
+ *
+ * @param {Object} task
+ * @param {Object} runInfo
+ * @param {string} category
+ * @returns {string}
+ */
+export function buildDoctorPrompt(task, runInfo, category) {
+  const currentCode = String(task.prompt || '').slice(0, 4000);
+  const errorText = [
+    String(runInfo.error || ''),
+    String(runInfo.stderr || ''),
+    String(runInfo.output || '').slice(0, 2000),
+  ].filter(Boolean).join('\n---\n');
+
+  return [
+    `Task: ${task.title || 'Task'} (type: ${task.type || 'prompt'})`,
+    `Schedule: ${task.scheduleText || task.schedule || ''}`,
+    `Failure Category: ${category}`,
+    `Exit Code: ${runInfo.exitCode ?? 'N/A'}`,
+    '',
+    'Execution Error & Traces:',
+    '```',
+    errorText || '(no error output)',
+    '```',
+    '',
+    'Current Code / Prompt:',
+    '```',
+    currentCode,
+    '```',
+    '',
+    'Instructions:',
+    '1. Explain what caused the failure in 1-2 concise sentences.',
+    '2. Provide the complete fixed code/prompt in `suggestedFix`. Make sure it preserves existing logic and fixes the specific failure.',
+    '3. Provide confidence level: "high", "medium", or "low".',
+  ].join('\n');
+}
+
+const DOCTOR_SYSTEM_PROMPT = [
+  'You are the autonomous Self-Healing Doctor for background scheduled tasks.',
+  'Analyze the failure and output ONLY a JSON object with this exact schema:',
+  '{',
+  '  "cause": "short 1-2 sentence description of root cause",',
+  '  "category": "syntax_error"|"dependency_missing"|"rate_limit"|"auth_failed"|"schema_change"|"timeout"|"network_error"|"runtime_error"|"validation_failed",',
+  '  "suggestedFix": "the exact new code or prompt that completely replaces the current code and resolves the issue",',
+  '  "confidence": "high"|"medium"|"low",',
+  '  "explanation": "brief explanation of what was fixed"',
+  '}',
+  'Do not include markdown fences outside the JSON. Return only the JSON.',
+].join(' ');
+
+/**
+ * Diagnoses a failed run and produces a doctor report.
+ * Uses deterministic analysis if no model is available, or model-assisted analysis when ask is supplied.
+ *
+ * @param {Object} params
+ * @param {Object} params.task Current task
+ * @param {Object} params.runInfo Run details
+ * @param {Function} [params.ask] Model querying function (optional)
+ * @param {string} [params.preferredModel] Preferred LLM model
+ * @param {number} [params.timeoutMs] Query timeout in ms
+ * @returns {Promise<Object>} Doctor diagnosis report
+ */
+export async function diagnoseFailure({ task, runInfo, ask, preferredModel, timeoutMs = 25000 }) {
+  if (!task) return { ok: false, error: 'No task provided' };
+
+  const currentCode = String(task.prompt || '');
+  const { category } = classifyFailure(task, runInfo);
+
+  // If a model querying function is available, ask the model for diagnosis and fix
+  if (typeof ask === 'function') {
+    try {
+      const prompt = buildDoctorPrompt(task, runInfo, category);
+      const res = await ask({
+        task,
+        prompt,
+        system: DOCTOR_SYSTEM_PROMPT,
+        preferredModel,
+        maxTokens: 1200,
+        timeoutMs,
+      });
+
+      if (res && res.ok && res.text) {
+        const parsed = parseJsonAnswer(res.text);
+        if (parsed && (parsed.suggestedFix || parsed.cause)) {
+          const suggestedFix = String(parsed.suggestedFix || currentCode);
+          const diff = generateUnifiedDiff(currentCode, suggestedFix, task.type === 'script' ? 'script.sh' : 'prompt.txt');
+          const confidence = ['high', 'medium', 'low'].includes(String(parsed.confidence).toLowerCase())
+            ? String(parsed.confidence).toLowerCase()
+            : 'medium';
+
+          return {
+            ok: true,
+            category: parsed.category || category,
+            cause: String(parsed.cause || 'Task encountered an unhandled execution failure').slice(0, 1000),
+            explanation: String(parsed.explanation || '').slice(0, 1000),
+            suggestedFix,
+            diff,
+            confidence,
+            timestamp: Date.now(),
+            applied: false,
+            autoHealed: false,
+          };
+        }
+      }
+    } catch {
+      // Fall through to deterministic rules
+    }
+  }
+
+  // Deterministic fallback heuristics when LLM is unavailable or unparseable
+  let cause = `Task failed with ${category}`;
+  let suggestedFix = currentCode;
+  let confidence = 'low';
+
+  if (category === FAILURE_CATEGORIES.TIMEOUT) {
+    cause = 'Execution exceeded timeout limit. Suggested increasing timeout or adding fast-fail flags.';
+    confidence = 'medium';
+  } else if (category === FAILURE_CATEGORIES.RATE_LIMIT) {
+    cause = 'Rate limit (429) hit. Suggested adding backoff delay between requests.';
+    confidence = 'medium';
+  } else if (category === FAILURE_CATEGORIES.SYNTAX_ERROR) {
+    cause = 'Syntax error detected in script or prompt syntax.';
+    confidence = 'low';
+  } else if (category === FAILURE_CATEGORIES.AUTH_FAILED) {
+    cause = 'Authentication failed (401/403). Required credentials or tokens are invalid/expired.';
+    confidence = 'medium';
+  }
+
+  const diff = generateUnifiedDiff(currentCode, suggestedFix);
+
+  return {
+    ok: true,
+    category,
+    cause,
+    explanation: 'Generated via deterministic heuristics',
+    suggestedFix,
+    diff,
+    confidence,
+    timestamp: Date.now(),
+    applied: false,
+    autoHealed: false,
+  };
+}
+
+/**
+ * Checks whether a task should be automatically healed by the Doctor.
+ * Enforces safe boundaries:
+ * 1. `task.autoHeal === true` or `task.selfHealing?.autoApply === true`
+ * 2. High confidence requirement
+ * 3. Daily quota: at most 1 auto-patch per 24 hours to prevent loops
+ *
+ * @param {Object} task
+ * @param {Object} doctorReport
+ * @returns {boolean}
+ */
+export function shouldAutoHeal(task, doctorReport) {
+  if (!task || !doctorReport) return false;
+
+  const isAutoEnabled = Boolean(task.autoHeal || task.selfHealing?.autoApply);
+  if (!isAutoEnabled) return false;
+
+  if (doctorReport.confidence !== 'high') return false;
+
+  const lastPatchedAt = task.selfHealing?.lastPatchedAt || 0;
+  const oneDayMs = 24 * 60 * 60 * 1000;
+  if (Date.now() - lastPatchedAt < oneDayMs) {
+    return false; // Daily patch limit reached
+  }
+
+  // Don't auto-patch if fix is identical
+  if (doctorReport.suggestedFix === task.prompt) {
+    return false;
+  }
+
+  return true;
+}
+
+/**
+ * Applies a doctor fix to the task in store, creating backups and updating revision history.
+ *
+ * @param {Object} task
+ * @param {Object} doctorReport
+ * @param {Object} options
+ * @param {Object} options.store Store instance
+ * @param {string} [options.user] Applying user / system identity
+ * @param {boolean} [options.isAutoHeal] Whether applied automatically
+ * @returns {Promise<{ ok: boolean, error?: string, revision?: Object }>}
+ */
+export async function applyDoctorFix(task, doctorReport, { store, user = 'operator', isAutoHeal = false } = {}) {
+  if (!task || !doctorReport || !doctorReport.suggestedFix) {
+    return { ok: false, error: 'Invalid task or doctor report' };
+  }
+
+  const previousCode = String(task.prompt || '');
+  const newCode = String(doctorReport.suggestedFix);
+
+  const revision = {
+    timestamp: Date.now(),
+    author: isAutoHeal ? 'doctor:auto' : `user:${user}`,
+    category: doctorReport.category || 'unknown',
+    cause: doctorReport.cause || '',
+    previousCode,
+    newCode,
+  };
+
+  const revisions = Array.isArray(task.revisions) ? [...task.revisions] : [];
+  revisions.unshift(revision);
+  if (revisions.length > 10) revisions.length = 10; // Keep up to 10 historical revisions
+
+  const updates = {
+    prompt: newCode,
+    previousPrompt: previousCode,
+    revisions,
+    updatedAt: Date.now(),
+  };
+
+  if (isAutoHeal) {
+    updates.selfHealing = {
+      ...(task.selfHealing || {}),
+      lastPatchedAt: Date.now(),
+      autoPatchesCount: ((task.selfHealing && task.selfHealing.autoPatchesCount) || 0) + 1,
+    };
+  }
+
+  // Update in store
+  if (store && typeof store.updateTask === 'function') {
+    await store.updateTask(task.id, updates);
+  } else if (store && store.tasks && store.tasks[task.id]) {
+    Object.assign(store.tasks[task.id], updates);
+  }
+
+  Object.assign(task, updates);
+  doctorReport.applied = true;
+  if (isAutoHeal) {
+    doctorReport.autoHealed = true;
+  }
+
+  return { ok: true, revision };
+}
+
+/**
+ * Reverts a task to its previous prompt/code from the last revision.
+ *
+ * @param {Object} task
+ * @param {Object} options
+ * @param {Object} options.store Store instance
+ * @param {string} [options.user] Reverting user identity
+ * @returns {Promise<{ ok: boolean, error?: string, revertedCode?: string }>}
+ */
+export async function revertDoctorFix(task, { store, user = 'operator' } = {}) {
+  if (!task) return { ok: false, error: 'No task provided' };
+
+  let revertCode = null;
+  const revisions = Array.isArray(task.revisions) ? [...task.revisions] : [];
+
+  if (revisions.length > 0) {
+    const lastRev = revisions[0];
+    revertCode = lastRev.previousCode;
+  } else if (task.previousPrompt) {
+    revertCode = task.previousPrompt;
+  }
+
+  if (revertCode === null || revertCode === undefined) {
+    return { ok: false, error: 'No previous revision or backup found to revert to' };
+  }
+
+  const currentCode = String(task.prompt || '');
+  const revertRevision = {
+    timestamp: Date.now(),
+    author: `revert:${user}`,
+    category: 'revert',
+    cause: 'Reverted to previous version',
+    previousCode: currentCode,
+    newCode: revertCode,
+  };
+
+  revisions.unshift(revertRevision);
+  if (revisions.length > 10) revisions.length = 10;
+
+  const updates = {
+    prompt: revertCode,
+    previousPrompt: currentCode,
+    revisions,
+    updatedAt: Date.now(),
+  };
+
+  if (store && typeof store.updateTask === 'function') {
+    await store.updateTask(task.id, updates);
+  } else if (store && store.tasks && store.tasks[task.id]) {
+    Object.assign(store.tasks[task.id], updates);
+  }
+
+  Object.assign(task, updates);
+  return { ok: true, revertedCode: revertCode };
+}
+