other
- Location
SKILL.md:87- Finding
Guidance Encourages Circumvention of CAPTCHA-Based Anti-Automation Controls
- Content
View full analysis
Vulnerability Details
File Location:
SKILL.md:87-97
Vulnerability Type: Anti-automation control evasion
Risk Level: MediumVulnerable Code Snippet:
python # Search URL Template url = f"https://scholar.google.com/scholar?q={research_area}+{target_company}" # Parse Logic # 1. Extract paper titles, authors, and citation counts # 2. Open author links and visit profile pages # 3. Extract H-index, total citations, and paper lists # 4. Locate the author's homepage link # Error Handling # - When a CAPTCHA is encountered: pause for 5 minutes and use a proxy IP # - If no results are returned: simplify the search terms # - If the page structure changes: verify the CSS selectorsTechnical Analysis
The Skill instructs the Agent to switch to a proxy IP after Google Scholar presents a CAPTCHA. A CAPTCHA is an anti-automation control intended to stop or challenge automated access. Changing network identity and resuming collection can bypass the practical effect of that control rather than respecting it.
This behavior is not necessary for the declared recruiting functionality. Candidate research can instead use approved APIs, manual browsing, cached results, or an explicit human verification step. The instruction also conflicts with the Skill's later claim that it will not bypass platform authentication mechanisms.
Attack Path
- A user requests automated candidate sourcing through Google Scholar.
- The Agent repeatedly loads Scholar search and profile pages.
- Google Scholar detects automated behavior and presents a CAPTCHA.
- Following
SKILL.md, the Agent pauses briefly and switches to a proxy IP. - The Agent resumes automated collection from a new network identity.
- Repeating this process may allow collection to continue despite the platform's anti-automation controls.
Impact Assessment
This does not directly grant local system privileges or access to private a ...[truncated 434 chars]
- Remediation
View remediation
Remediation Suggestions
- Remove all instructions recommending proxy rotation or network-identity changes in response to a CAPTCHA.
- Require automation to stop immediately when a CAPTCHA or equivalent access challenge appears.
- Require explicit human interaction before continuing, where permitted by the platform.
- Prefer approved APIs and authenticated integrations with documented rate limits.
- Add per-platform request throttling, exponential backoff, caching, and a maximum request budget.
- Add an allowlist of approved platforms and collection methods.
- Record the data source, collection time, legal basis, and consent or legitimate-interest basis where applicable.
- Ensure the operational instructions are consistent with the Skill's stated prohibition on bypassing platform controls.