Skip to content
Back to skills

Human Captcha Intervention

ASecurity

'Browser-based automation encounters a CAPTCHA challenge that cannot

  • 2 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added September 19, 2026
ai-agentsbash

Works with

  • cli

Security analysis

A100/100

Scanned September 19, 2026

npx -y skills add Stranglehold/Exocortex --skill human-captcha-intervention --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Human Captcha Intervention?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Human Captcha Intervention
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/stranglehold-human-captcha-intervention/badge)](https://www.skillsdirectory.com/skills/stranglehold-human-captcha-intervention)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: human-captcha-intervention
description: 'Browser-based automation encounters a CAPTCHA challenge that cannot
  be solved programmatically. Keywords: "CAPTCHA",...'
triggers:
- 'Browser-based automation encounters a CAPTCHA challenge that cannot be solved programmatically.
  Keywords: "CAPTCHA",...'
version: '1.0'
author: Exocortex
---

# Skill: Human CAPTCHA Intervention

## Trigger
Browser-based automation encounters a CAPTCHA challenge that cannot be solved programmatically. Keywords: "CAPTCHA", "verify you're human", "checkbox verification", "image selection puzzle", "reCAPTCHA", "hCaptcha", "Cloudflare challenge", "browser stuck on verification".

## Inputs Required
- **Page URL** — where the CAPTCHA is blocking progress
- **CAPTCHA type** — checkbox, image selection, text entry, sliding puzzle, etc.
- **Screenshot path** — visual evidence of the challenge (e.g., `/a0/usr/chats/.../browser/screenshots/*.png`)
- **Task context** — what was being attempted when blocked

## Procedure

### Phase 1: Confirm CAPTCHA Blockage
Before requesting human intervention, verify it's actually a CAPTCHA and not another issue.

1. **Check the page title and visible text:**
   ```bash
   # Look for common CAPTCHA indicators in page content
   grep -i "captcha\|verify.*human\|checkbox\|select all" /path/to/page/source.html 2>/dev/null || echo "No obvious CAPTCHA markers"
   ```

2. **Verify screenshot exists and shows the challenge:**
   ```bash
   ls -la /a0/usr/chats/*/browser/screenshots/*.png | tail -3
   ```

3. **Confirm browser agent hit step limit or explicit error:**
   Check for messages like "Task reached step limit" or "CAPTCHA detected".

### Phase 2: Present CAPTCHA to Human Operator
Format the request clearly so the human can solve it efficiently.

1. **Display the screenshot inline:**
   ```markdown
   ![CAPTCHA Challenge](/a0/usr/chats/<chat_id>/browser/screenshots/<screenshot>.png)
   ```

2. **Provide context in a structured format:**
   ```markdown
   ## CAPTCHA Intervention Required

   **URL:** https://example.com/signup
   **Type:** [ ] reCAPTCHA checkbox  [ ] Image selection  [ ] Text entry  [ ] Other: ___
   **Task blocked:** Email signup for X.com registration
   
   **Instructions:** Please solve the CAPTCHA shown above and provide:
   - If checkbox: Confirm you clicked it
   - If image selection: Describe which images to select (e.g., "traffic lights", "crosswalks")
   - If text entry: Provide the exact characters shown
   ```

3. **Wait for human response** — Do not proceed until operator provides solution.

### Phase 3: Execute Human Solution
Once the operator provides the answer, translate it into browser actions.

1. **For checkbox CAPTCHAs:**
   ```json
   {
     "tool_name": "browser_agent",
     "tool_args": {
       "message": "Click the 'I'm not a robot' checkbox to complete verification, then continue with [original task]."
     }
   }
   ```

2. **For image selection CAPTCHAs:**
   ```json
   {
     "tool_name": "browser_agent",
     "tool_args": {
       "message": "Select all images containing [operator's answer, e.g., 'traffic lights'], then click SUBMIT to verify."
     }
   }
   ```

3. **For text entry CAPTCHAs:**
   ```json
   {
     "tool_name": "browser_agent",
     "tool_args": {
       "message": "Type '[operator's answer]' into the verification field, then submit and continue with [original task]."
     }
   }
   ```

### Phase 4: Verify Success and Resume
Confirm the CAPTCHA was solved and the original task can proceed.

1. **Check for success indicators:**
   - Page navigates forward (URL changes)
   - Form becomes active/submittable
   - No error messages about verification

2. **Resume original task** with updated instructions if needed.

## Quality Checks
- [ ] CAPTCHA actually confirmed before requesting human help (not just step limit hit for other reasons)
- [ ] Screenshot path is correct and accessible to operator
- [ ] CAPTCHA type clearly identified in request to human
- [ ] Human solution translated correctly into browser actions
- [ ] Success verified before resuming original task

## Anti-Patterns
- **Requesting human help without confirming it's a CAPTCHA.** Step limits can be hit for many reasons — slow pages, complex forms, navigation issues. Verify first.
- **Not providing the screenshot.** Humans need to see what they're solving. Always include the image path or embed it.
- **Vague instructions to humans.** "Solve this" is useless. Specify exactly what type of CAPTCHA and what information you need back.
- **Resuming before verification.** If you assume success without checking, you'll loop on the same CAPTCHA again.

## Example Interaction

**Agent:**
```
![CAPTCHA Challenge](/a0/usr/chats/Gg4aZsTw/browser/screenshots/QAefyr6q.png)

## CAPTCHA Intervention Required

**URL:** https://proton.me/mail/signup  
**Type:** reCAPTCHA checkbox ("I'm not a robot")
**Task blocked:** ProtonMail account creation for X.com registration

Please confirm you want me to click the "I'm not a robot" checkbox, or provide alternative instructions.
```

**Human:**
```
Yes, click the checkbox and continue with signup using mountain_7620@proton.me
```

**Agent:** (executes browser_agent with checkbox instruction)

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.