Back to skill

Security audit

Scrapling Official Skill

Security checks for vulnerabilities and agentic risk

Overview

The skill coherently supports web scraping, with disclosed optional browser sessions, saved login state, and proxies that need careful handling.

Prefer a dedicated browser context. Treat saved login files and captured traffic as credentials: keep them private, exclude them from commits and uploads, and remove them when finished. Choose trusted proxies rather than random sponsors, and authorize any requests that modify remote data. Pin dependencies for repeatable installations. The review independently checked the scanner's material gaps; installed package and container behavior remains outside the inspected artifacts.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
Findings (35)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Although the description mentions HTTP requests, its stated purpose is scraping and data extraction. The code instead invokes Fetcher.delete repeatedly, requesting removal of a server-side resource. Whether deletion succeeds depends on the target server, but the attempted state-changing operation is materially outside the declared scraping purpose. The header and proxy options are supporting implementation details, not separate mismatches.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied chunk exclusively demonstrates PUT requests, including sending status=updated to an /update endpoint. Remote-state mutation is materially different from the declared scraping and extraction purpose. Browser impersonation and proxy configuration are compatible supporting details and are not independently mismatches. The examples show write-capable requests, though they do not establish that any remote update actually succeeds.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The declared purpose describes data retrieval and extraction, but the supplied code uses AsyncFetcher.delete to request deletion of a resource. This is a state-changing capability not disclosed by the scraping description. The example URL does not establish that any resource is actually deleted, but the requested operation is still materially different from scraping. Stealth headers, impersonation, and proxy routing are supporting implementation details, not separate mismatches.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The declared purpose covers scraping and supporting HTTP requests, but this chunk specifically demonstrates PUT-based updates, including a status update sent to an /update endpoint. That is a write capability not disclosed by the scraping-focused description. The code does not establish that any remote state actually changes. Stealth headers, impersonation, and proxy configuration are supporting implementation details, not separate mismatches.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The supplied code performs document indexing for a retrieval system rather than website scraping or crawling. Although scraped content could feed this workflow, vector-store ingestion is a distinct capability not described in the declared purpose. The snippet does not show network fetching, browser automation, or Scrapling usage.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

GET requests, saving fetched content, and stealth-browser fetching align with scraping. POST can also support scraping workflows. However, exposing general PUT and DELETE operations adds remote resource modification and deletion capabilities beyond the declared data-extraction purpose. The chunk documents these commands but does not show their implementation or establish that any modifying requests were executed.

Content

No source excerpt is available for this finding.

YARA rule 'info_stealer': Information stealer patterns (credential harvesting, browser data theft) [malware]

High
Category
YARA Match
Confidence
75% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

Scanner excerpt · SKILL.md (reported line 262)May include surrounding context.

md
When reusing a context, `useragent`, `locale`, `timezone_id`, `proxy`, `proxy_rotator` and `additional_args` are ignored. Per-request `proxy` is also ignored. The existing context settings remain in use.

Scrapling creates its own tabs in either mode and closes only those tabs. Existing tabs stay open. Page options such as `extra_headers`, `timeout`, `page_setup` and `page_action` still work.

`cookies` and `init_script` can still be added, but they affect the shared context when `new_context=False`. The remote classes do not accept `real_chrome`; `cdp_url` selects the browser. Other browser launch options do not apply because the browser is already running.

### Shadow DOM

Use `pierce_shadow=True` with StealthyFetcher and its session classes to include open shadow roots. It defaults to `False` and can be overridden per request. For MCP, pass `pierce_shadow=true` to `browser_fetch_once` or each `browser_fetch` call.

The browser CLI command accepts `--pierce-shadow` and `--no-pierce

YARA rule 'info_stealer': Information stealer patterns (credential harvesting, browser data theft) [malware]

High
Category
YARA Match
Confidence
75% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

Scanner excerpt · references/fetching/stealthy.md (reported line 62)May include surrounding context.

When reusing a context, useragent, locale, timezone_id, proxy, proxy_rotator and additional_args are ignored. Per-request proxy is also ignored. The existing context settings remain in use.

Scrapling creates its own tabs in either mode and closes only those tabs. Existing tabs stay open. Page options such as extra_headers, timeout, page_setup and page_action still work.

cookies and init_script can still be added, but they affect the shared context when new_context=False. The remote classes do not accept real_chrome; cdp_url selects the browser. Other browser launch options do not apply because the browser is already running.

Use AsyncRemoteSession with async with and await:

python
from scrapling.fetchers import AsyncRemoteSession

async def scrape_remote():
    async with AsyncRemoteSession(cdp_url="http://localhost:9222", new_context=False) as session:
        return await session.fetch("https://example.com")

What does it do?

St

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 274)May include surrounding context.

md
nt selectors, or include those elements in direct-child paths. See [Shadow DOM](references/fetching/stealthy.md#shadow-dom) for assigned content, hidden slots,

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 425)May include surrounding context.

md
local after closure; unfinished captures are not awaited. See [Migrating in v5](references/fetching/stealthy.md#migrating-in-v5) for the removed API capture int

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 429)May include surrounding context.

md
- `references/mcp-server.md` - MCP server tools, persistent session management, remote browsers over CDP, authentication, and capabilities

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 434)May include surrounding context.

md
- `references/integrations/scrapy.md` - Using Scrapling's parsing API inside existing Scrapy projects through the `scrapling_response` decorator

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: network.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Docker image referenced without tag or digest: 'docker pull pyd4vinci/scrapling'.

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The untagged image implicitly resolves to a mutable latest tag, so the recommended installation is not bound to reviewed image contents. If the published image changes maliciously or is compromised, subsequent execution could expose mounted files or credentials; pulling alone does not execute the image.

Content

No source excerpt is available for this finding.

Docker image referenced without tag or digest: 'docker pull ghcr.io/d4vinci/scrapling:latest'.

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

Although this reference includes a tag, latest is mutable and does not identify fixed image contents. A compromised or unexpected replacement could execute with the container's granted access when users subsequently run it.

Content

No source excerpt is available for this finding.

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
60% confidence
Finding

Tool defaults are unsafe or overly permissive (e.g. disabled TLS verification, no authentication, world-writable permissions). Unsafe defaults widen the attack surface.

Content

Scanner excerpt · SKILL.md (reported line 329)May include surrounding context.

text
Press Ctrl+C to pause gracefully - pending requests and seen URLs are saved automatically. Later, when you start the spider again, pass the same `crawldir`, and it will resume from where it stopped.

While iterating on a spider's `parse()` logic, set `development_mode = True` on the spider class to cache responses to disk on the first run and replay them on subsequent runs - so you can re-run the spider as many times as you want without re-hitting the target servers. The cache lives in `.scrapling_cache/{spider.name}/` by default and can be overridden with `development_cache_dir`. Don't ship a spider with this enabled.

For rules-based crawls (follow links matching a regex), use `CrawlSpider` instead of writing the link-extraction loop yourself:
```python

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
60% confidence
Finding

Tool defaults are unsafe or overly permissive (e.g. disabled TLS verification, no authentication, world-writable permissions). Unsafe defaults widen the attack surface.

Content

Scanner excerpt · references/spiders/advanced.md (reported line 165)May include surrounding context.

When you're iterating on a spider's parse() logic, re-hitting the target servers on every run is slow and noisy. Development mode caches every response to disk on the first run and replays them from disk on subsequent runs, so you can tweak your selectors and re-run the spider as many times as you want without fetching those pages again.

Enable it by setting development_mode = True on your spider:

python
class MySpider(Spider):

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
60% confidence
Finding

Tool defaults are unsafe or overly permissive (e.g. disabled TLS verification, no authentication, world-writable permissions). Unsafe defaults widen the attack surface.

Content

Scanner excerpt · references/spiders/advanced.md (reported line 171)May include surrounding context.

md
class MySpider(Spider):
    name = "my_spider"
    start_urls = ["https://example.com"]
    development_mode = True

    async def parse(self, response: Response):
        yield {"title": response.css("title::text").get("")}

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · references/fetching/choosing.md (reported line 19)May include surrounding context.

md
| Feature            | Fetcher                                           | StealthyFetcher                                                                            |
|--------------------|---------------------------------------------------|--------------------------------------------------------------------------------------------|
| Relative speed     | 🐇🐇🐇🐇🐇                                        | 🐇🐇🐇                                                                                     |
| Stealth            | ⭐⭐                                                | ⭐⭐⭐⭐⭐                                                                                      |
| Anti-Bot options   | ⭐⭐                                                | ⭐⭐⭐⭐⭐                                                                                      |
| JavaScript loading | ❌                                                 | ✅                                                                                          |

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/fetching/static.md (reported line 96)May include surrounding context.

python
page.css('.something.something')

page = Fetcher.get('https://api.github.com/events')
>>> page.json()
[{'id': '<redacted>',
  'type': 'PushEvent',

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/fetching/static.md (reported line 104)May include surrounding context.

md
'login': '<redacted>',
   'display_login': '<redacted>',
   'gravatar_id': '',
   'url': 'https://api.github.com/users/<redacted>',
   'avatar_url': 'https://avatars.githubusercontent.com/u/<redacted>'},
  'repo': {'id': '<redacted>',
...

The examples include remote-resource updates and deletion, capabilities outside the stated scraping and extraction purpose.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

The examples at L137–L149 send PUT requests with an explicitly updated status, and L157–L166 issue DELETE requests against a resource endpoint. Unlike fetching pages or submitting authentication forms needed for scraping, these examples demonstrate modifying and deleting remote resources without an extraction-related justification.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/fetching/static.md (reported line 189)May include surrounding context.

Make multiple requests with the same settings and the same cookies

text
page1 = session.get('https://scrapling.requestcatcher.com/get')
page2 = session.post('https://scrapling.requestcatcher.com/post', data={'key': 'value'})
page3 = session.get('https://api.github.com/events')

# All requests share the same session and connection pool
text

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · references/fetching/stealthy.md (reported line 115)May include surrounding context.

md
|    ✔️    |
|      extra_flags      | A list of additional browser flags to pass to the browser on launch.                                                                                                                                                                |    ✔️    |
|   solve_cloudflare    | When enabled, fetcher solves all types of Cloudflare's Turnstile/Interstitial challenges before returning the response to you. Defaults to `False`.                                                                                                      |    ✔️    |
|     block_webrtc      | Forces WebRTC to respect proxy settings to prevent local IP address leak.                                                                                                                                                           |    ✔️    |
|      hide_canvas      | Add random noise to canvas operations to prevent fingerprinting.

Session-state export instructions omit a warning that the saved JSON can contain sensitive authentication data.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The instructions encourage saving post-login cookies, local storage, and IndexedDB to a JSON file without guidance for protecting authentication material. In an agent workspace, accidental commits, shared artifacts, or permissive file access could expose reusable sessions and enable account impersonation; this is unsafe handling guidance rather than evidence of deliberate theft.

Content

No source excerpt is available for this finding.