Back to skill

Security audit

ProxyLane Web Evidence

Security checks for vulnerabilities and agentic risk

Overview

This skill is a bounded web evidence collector that uses user-supplied proxy/session inputs and writes redacted receipts; I found no hidden, destructive, or purpose-incompatible behavior.

Install only in an isolated environment, provide proxy and MCP secrets through approved private channels, set the intended language/country options explicitly for Google runs, and review the pinned dependencies before live collection. Treat outputs as bounded observations with documented gaps, not proof of complete access, regional causality, representative sampling, or checkout-verified pricing.

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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (17)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared purpose centers on collecting external web observations through ProxyLane or MCP from specific online sources. The actual code operates entirely on local repository files: it reads files, searches for secrets and credential-like strings, builds a manifest, creates a zip package, and outputs a receipt. There is no network access, no use of ProxyLane, no HTTP requests, no MCP session usage, and no logic for YouTube, Google AI, Xiaohongshu, or product-price observation. This is a materially different primary purpose, so the description does not accurately represent the code.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

This is a clear description-behavior mismatch. The declared purpose is a web-observation/collection skill using external connections and specific source types. The supplied code instead implements a local trigger-routing validator that reads a JSON proposal file and returns whether the provided text exactly matches one of the configured resolver triggers. Its primary purpose is route validation, not web evidence collection. It does not access the network, use ProxyLane, interact with MCP, or retrieve any source content from YouTube, Google AI, Xiaohongshu, or product pages.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description is about web data collection through ProxyLane and MCP-backed source observation. The actual code shown does not perform web collection, proxy access, source tracing, citation retrieval, YouTube/Xiaohongshu/product observation, or diagnostics against external sources. Instead, it tests a packaging utility focused on secure packaging and scanning of files for secrets and unsafe URLs. This is a materially different primary purpose and capability set, so the description does not accurately represent the supplied code chunk.

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 59)May include surrounding context.

md
3. **Run a fixture demonstration first when evaluating the skill.** `python3 scripts/collect.py --mode demo --fixture --out output/fixture-demo`. This calls all

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
90% confidence
Finding

The skill declares powerful tools (exec, read, write) and the analyzer detected additional effective capabilities including environment access, network, shell, and MCP interaction, but the manifest does not define a restrictive tool scope such as permissions or allowed-tools. This creates an overly broad execution boundary: an agent invoking the skill may be able to read local files, access secrets in the environment, write artifacts, and make network calls beyond what is minimally required for bounded evidence collection.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This JSON fixture includes natural-language/request parameters that force a specific locale via requested_country: "us" and requested_language: "en" in the Google search context. Because the file does not indicate user opt-in or a documented region-specific justification, it conflicts with the policy against imposing a language/locale without choice.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: requests==2.32.5 — 2 advisory(ies): CVE-2026-25645 (Requests has Insecure Temp File Reuse in its extract_zipped_paths() utility func); CVE-2026-25645 (Requests is a HTTP library. Prior to version 2.33.0, the `requests.utils.extract)

Medium
Category
Supply Chain
Confidence
95% confidence
Finding

The pinned dependency requests==2.32.5 is flagged with a known vulnerability affecting extract_zipped_paths() in versions prior to 2.33.0. Even if this skill primarily performs HTTP collection and may not directly call the vulnerable utility, shipping a known-vulnerable version unnecessarily exposes any code path or future change that uses that functionality, and dependency issues in a web-facing evidence-collection skill increase risk because it processes external content and URLs.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/collect.py (reported line 254)May include surrounding context.

python
try:
            completed = None
            for attempt in range(args.attempts):
                completed = subprocess.run(command, capture_output=True, text=True, timeout=max(10, args.timeout - 5))
                if completed.returncode == 0:
                    break
                if attempt + 1 < args.attempts:

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/collect.py (reported line 697)May include surrounding context.

python
try:
            with tempfile.TemporaryDirectory(prefix="proxylane-evidence-") as scratch:
                # POSIX process group also owns browser/yt-dlp/JS descendants.
                child = subprocess.Popen([sys.executable, str(Path(__file__).resolve()), *sys.argv[1:], "--_worker", "--_scratch", scratch], stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, start_new_session=True)
                try:
                    stdout, _ = child.communicate(timeout=args.timeout)
                    result = json.loads(stdout)

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
75% 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 · tests/test_integration.py (reported line 24)May include surrounding context.

python
owner = self
        self.calls = []
        self.logged_in = True
        self.reject_auth = False
        self.sse = False
        self.missing_tools = False
        self.raw = c.fixture("xiaohongshu")

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · tests/test_integration.py (reported line 117)May include surrounding context.

python
def run_cli(self, *arguments, environment=None):
        command = [sys.executable, str(ROOT / "scripts/collect.py"), *arguments]
        result = subprocess.run(command, capture_output=True, text=True, env=environment or self.environment(), timeout=25)
        return result, json.loads(result.stdout)

    def test_all_modes_stdout_artifact_contract_and_fixture_boundary(self):

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · tests/test_integration.py (reported line 141)May include surrounding context.

python
node = shutil.which("node")
        if not node:
            self.skipTest("Node bridge requires an installed Node executable")
        done = subprocess.run([node, str(ROOT / "scripts/proxylane-web-evidence.mjs"), "--mode", "demo", "--fixture"], capture_output=True, text=True, env=self.environment(), timeout=25)
        self.assertEqual(done.returncode, 0)
        self.assertEqual(json.loads(done.stdout)["status"], "fixture_demo")

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · tests/test_integration.py (reported line 160)May include surrounding context.

python
self.assertLess(time.monotonic() - started, 18)
            self.assertTrue(pidfile.exists())
            pid = pidfile.read_text().strip()
            state = subprocess.run(["ps", "-p", pid, "-o", "stat="], capture_output=True, text=True).stdout.strip()
            self.assertTrue(not state or state.startswith("Z"), "Source descendant remains running")
            self.assertFalse(Path(configfile.read_text()).exists())
            self.assertNotIn("fixture-password", result.stdout)

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · tests/test_routing.py (reported line 32)May include surrounding context.

python
proposal = json.loads((ROOT / "references/routing-proposal.json").read_text())
        self.assertEqual(triggers, proposal["resolver_triggers"])
        self.assertIn("name: " + proposal["name"], skill)
        done = subprocess.run([sys.executable, str(ROOT / "scripts/route.py"), "CAPTURE GOOGLE AI CITATIONS FOR A REGIONAL COMPARISON!"], capture_output=True, text=True, timeout=5)
        self.assertEqual(done.returncode, 0)
        self.assertEqual(json.loads(done.stdout)["route"], "proxylane-web-evidence")

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · tests/test_routing.py (reported line 41)May include surrounding context.

python
phrase = "fetch YouTube transcripts through my proxy for RAG"
        self.assertEqual(route.resolve(phrase)["route"], "proxylane-web-evidence")
        with tempfile.TemporaryDirectory() as temporary:
            done = subprocess.run([sys.executable, str(ROOT / "scripts/collect.py"), "--mode", "youtube-transcript", "--fixture", "--out", temporary], capture_output=True, text=True, timeout=10)
            self.assertEqual(done.returncode, 0)
            outcome = json.loads((Path(temporary) / "outcome.json").read_text())
            self.assertEqual(outcome["proof_mode"], "fixture")

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

The parser sets default values for languages and locale-related options (--languages en, --country us, --language en, --locale en-US). This creates a default behavior that prefers a specific language/locale even when the user has not explicitly chosen one, which can violate the policy against forcing a locale without opt-in.

Content

No source excerpt is available for this finding.

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · tests/test_collect.py (reported line 191)May include surrounding context.

python
def inner_text(self): return self.text
    def locator(self, selector): return self.children[selector]
    def all(self): return self.anchors
    def get_attribute(self, key): return getattr(self, key, None)


class GoogleTests(Base):