fix: robustness improvements — atomic writes, timeouts, shell injection, validation errors

- Atomic JSON writes (write-to-tmp + rename) prevent queue/log corruption
- Per-job (3min) and overall run (45min) timeouts prevent hangs
- execFileSync in ai_answer.mjs prevents shell injection with resume paths
- Validation error detection after form fill in Easy Apply modal
- Config-driven enabled_apply_types (from settings.json)
- isRequired() detects required/aria-required/label * patterns
- getLabel() strips trailing * from required field labels
- Actionable logging on failures ("Action: ..." messages)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

lib/ai_answer.mjs

@@ -16,20 +16,25 @@ import { ANTHROPIC_API_URL } from './constants.mjs';
 export async function generateAnswer(question, profile, apiKey, job = {}) {
   if (!apiKey) return null;
 
-  // Read resume text if available
+  // Read resume text if available — try pdftotext for PDFs, fall back to raw read
   let resumeText = '';
   if (profile.resume_path && existsSync(profile.resume_path)) {
-    try {
-      // Try to read as text — PDF will be garbled but still useful for key facts
-      // If pdftotext is available, use it; otherwise skip
-      const { execSync } = await import('child_process');
+    // Only attempt pdftotext for .pdf files
+    if (profile.resume_path.toLowerCase().endsWith('.pdf')) {
       try {
-        resumeText = execSync(`pdftotext "${profile.resume_path}" -`, { timeout: 5000 }).toString().slice(0, 4000);
+        const { execFileSync } = await import('child_process');
+        // execFileSync avoids shell injection — args passed as array, not interpolated
+        resumeText = execFileSync('pdftotext', [profile.resume_path, '-'], { timeout: 3000 }).toString().slice(0, 4000);
       } catch {
-        // pdftotext not available — skip resume text
+        // pdftotext not available or failed — skip
+      }
+    } else {
+      // Plain text resume
+      try {
+        resumeText = readFileSync(profile.resume_path, 'utf8').slice(0, 4000);
+      } catch {
+        // ignore
       }
-    } catch {
-      // ignore
     }
   }
 

lib/apply/easy_apply.mjs

@@ -83,7 +83,9 @@ export async function apply(page, job, formFiller) {
       Array.from(document.querySelectorAll('[aria-label*="Easy Apply"], [aria-label*="Apply"]'))
         .map(el => ({ tag: el.tagName, aria: el.getAttribute('aria-label'), visible: el.offsetParent !== null }))
     ).catch(() => []);
-    console.log(`    ℹ️  No Easy Apply element found. Apply-related elements: ${JSON.stringify(applyEls)}`);
+    console.log(`    ℹ️  No Easy Apply button found. Page URL: ${page.url()}`);
+    console.log(`    ℹ️  Apply-related elements on page: ${JSON.stringify(applyEls)}`);
+    console.log(`    Action: job may have been removed, filled, or changed to external apply`);
     return { status: 'skipped_easy_apply_unsupported', meta };
   }
 
@@ -97,7 +99,11 @@ export async function apply(page, job, formFiller) {
   // Click Easy Apply and wait for modal to appear
   await page.click(LINKEDIN_APPLY_BUTTON_SELECTOR, { timeout: APPLY_CLICK_TIMEOUT }).catch(() => {});
   const modal = await page.waitForSelector(LINKEDIN_EASY_APPLY_MODAL_SELECTOR, { timeout: 8000 }).catch(() => null);
-  if (!modal) return { status: 'no_modal', meta };
+  if (!modal) {
+    console.log(`    ❌ Modal did not open after clicking Easy Apply`);
+    console.log(`    Action: LinkedIn may have changed the modal structure or login expired`);
+    return { status: 'no_modal', meta };
+  }
 
   const MODAL = LINKEDIN_EASY_APPLY_MODAL_SELECTOR;
 
@@ -146,6 +152,22 @@ export async function apply(page, job, formFiller) {
 
     await page.waitForTimeout(MODAL_STEP_WAIT);
 
+    // Check for validation errors after form fill — if LinkedIn shows errors,
+    // the form won't advance. Re-check errors AFTER fill since fill may have resolved them.
+    const postFillErrors = await page.evaluate((sel) => {
+      const modal = document.querySelector(sel);
+      if (!modal) return [];
+      return Array.from(modal.querySelectorAll('[class*="error"], [aria-invalid="true"], .artdeco-inline-feedback--error'))
+        .map(e => e.textContent?.trim().slice(0, 80)).filter(Boolean);
+    }, MODAL).catch(() => []);
+
+    if (postFillErrors.length > 0) {
+      console.log(`    [step ${step}] ❌ Validation errors after fill: ${JSON.stringify(postFillErrors)}`);
+      console.log(`    Action: check answers.json or profile.json for missing/wrong answers`);
+      await dismissModal(page, MODAL);
+      return { status: 'incomplete', meta, validation_errors: postFillErrors };
+    }
+
     // --- Button check order: Next → Review → Submit ---
     // Check Next first — only fall through to Submit when there's no forward navigation.
     // This prevents accidentally clicking a Submit-like element on early modal steps.
@@ -195,7 +217,8 @@ export async function apply(page, job, formFiller) {
       return { status: 'stuck', meta };
     }
 
-    console.log(`    [step ${step}] no Next/Review/Submit found — breaking`);
+    console.log(`    [step ${step}] ❌ No Next/Review/Submit button found in modal`);
+    console.log(`    Action: LinkedIn may have changed button text/structure. Check button snapshot above.`);
     break;
   }
 

lib/constants.mjs

@@ -87,3 +87,7 @@ export const EXTERNAL_ATS_PATTERNS = [
 
 // --- Queue ---
 export const DEFAULT_MAX_RETRIES = 2;
+
+// --- Run limits ---
+export const APPLY_RUN_TIMEOUT_MS = 45 * 60 * 1000; // 45 minutes
+export const PER_JOB_TIMEOUT_MS = 3 * 60 * 1000; // 3 minutes per job

lib/form_filler.mjs

@@ -119,10 +119,30 @@ export class FormFiller {
       const ariaLabel = node.getAttribute('aria-label') || '';
       const ariaLabelledBy = node.getAttribute('aria-labelledby');
       const linked = ariaLabelledBy ? document.getElementById(ariaLabelledBy)?.textContent?.trim() : '';
-      return forLabel || ariaLabel || linked || node.placeholder || node.name || '';
+      // Clean up — remove trailing * from required field labels
+      const raw = forLabel || ariaLabel || linked || node.placeholder || node.name || '';
+      return raw.replace(/\s*\*\s*$/, '').trim();
     }).catch(() => '');
   }
 
+  /**
+   * Check if a form element is required.
+   * LinkedIn uses multiple patterns: required attribute, aria-required, or * in label.
+   */
+  async isRequired(el) {
+    return await el.evaluate(node => {
+      if (node.required || node.getAttribute('required') !== null) return true;
+      if (node.getAttribute('aria-required') === 'true') return true;
+      // Check if the associated label contains *
+      const id = node.id;
+      if (id) {
+        const label = document.querySelector(`label[for="${id}"]`);
+        if (label && label.textContent.includes('*')) return true;
+      }
+      return false;
+    }).catch(() => false);
+  }
+
   /**
    * Select the first option from an autocomplete dropdown.
    * Waits for the dropdown to appear, then clicks the first option.
@@ -181,8 +201,7 @@ export class FormFiller {
           await this.selectAutocomplete(page, inp);
         }
       } else if (!answer) {
-        const required = await inp.getAttribute('required').catch(() => null);
-        if (required !== null) unknown.push(lbl);
+        if (await this.isRequired(inp)) unknown.push(lbl);
       }
     }
 
@@ -196,8 +215,7 @@ export class FormFiller {
       if (answer) {
         await ta.fill(answer).catch(() => {});
       } else {
-        const required = await ta.getAttribute('required').catch(() => null);
-        if (required !== null) unknown.push(lbl);
+        if (await this.isRequired(ta)) unknown.push(lbl);
       }
     }
 

lib/queue.mjs

@@ -3,7 +3,7 @@
  * Handles jobs_queue.json read/write/update
  * Uses in-memory cache to avoid redundant disk I/O within a run.
  */
-import { readFileSync, writeFileSync, existsSync, mkdirSync } from 'fs';
+import { readFileSync, writeFileSync, renameSync, existsSync, mkdirSync } from 'fs';
 import { dirname, resolve } from 'path';
 import { fileURLToPath } from 'url';
 
@@ -37,6 +37,17 @@ function ensureDir(path) {
   if (!existsSync(dir)) mkdirSync(dir, { recursive: true });
 }
 
+/**
+ * Atomic write — writes to a temp file then renames.
+ * Prevents corruption if two processes write simultaneously or the process
+ * crashes mid-write. rename() is atomic on POSIX filesystems.
+ */
+function atomicWriteJSON(filePath, data) {
+  const tmp = filePath + '.tmp';
+  writeFileSync(tmp, JSON.stringify(data, null, 2));
+  renameSync(tmp, filePath);
+}
+
 // --- In-memory caches (populated on first read, invalidated on write) ---
 let _queueCache = null;
 let _logCache = null;
@@ -50,7 +61,7 @@ export function loadQueue() {
 
 export function saveQueue(jobs) {
   ensureDir(QUEUE_PATH);
-  writeFileSync(QUEUE_PATH, JSON.stringify(jobs, null, 2));
+  atomicWriteJSON(QUEUE_PATH, jobs);
   _queueCache = jobs;
 }
 
@@ -62,7 +73,8 @@ function loadLog() {
 }
 
 function saveLog(log) {
-  writeFileSync(LOG_PATH, JSON.stringify(log, null, 2));
+  ensureDir(LOG_PATH);
+  atomicWriteJSON(LOG_PATH, log);
   _logCache = log;
 }