Back to skill

Security audit

signal-scout

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with medium confidence, mainly because it reports instructions to bypass host network restrictions and several capabilities or output boundaries that do not match the stated research purpose.

The Review verdict means scanners found meaningful concerns without establishing malicious intent. Before installing, consider whether the skill's network-bypass guidance and private-address option fit your needs, and whether you trust its stated purpose and output controls. Inspection was partial: one script could not be fully analyzed, so the scan does not establish that the artifact is safe.

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
Findings (1)

T01 · Skill Instruction Hijacking

Warning
Location
references/research-framework.md
Finding

Instructions explicitly bypass a host tool's domain restriction

Content
View full analysis

Impact

An agent following these instructions can access a domain that its host fetch tool deliberately restricts, without establishing that the alternate access path is authorized.

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
Findings (24)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied script has a materially different primary purpose: measuring repeat usage of the tool. Although it examines output files associated with the declared workflow, its report does not implement or directly support identifying and verifying prospects. This is an undeclared maintainer-facing analytics capability. The mismatch is limited to purpose and functionality; the code shows no network transmission, telemetry collection beyond local filename metadata, or file modification.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The mismatch is not simply that this helper does no prospect research itself; supporting modules need not implement the entire declared workflow. The material conflict is that its filtering functions deliberately eliminate people-to-message results, a core promised output. This could support an optional privacy mode, but that mode and its restrictions are absent from the declared description. The supplied code performs in-memory filtering and URL parsing; it shows no network access, exfiltration, or unrelated resource access.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
85% confidence
Finding

Most of this code is supporting implementation consistent with the declared purpose. The narrow mismatch is the explicit opt-in expansion of source verification to non-public network resources. Although documented as useful for checking a local page copy and disabled by default, that capability is not represented in the declared public-only scope. The verifier implementation is not supplied, so this finding concerns the capability explicitly exposed and delegated by this script, not proof of any particular private resource being fetched.

Content

No source excerpt is available for this finding.

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 155)May include surrounding context.

md
tiers in place and writing `handoff.json` next to the input) → HTML report (`generate_report.py`, with New / Seen Nx / Resurfacing badges auto-detected from sib

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 171)May include surrounding context.

md
e out: re-run with `--use-proxy`. The underlying scripts (`verify_sources.py`, `generate_report.py`, `diff_reports.py`) remain individually runnable when you ne

Reframes a tool's domain restriction as an obstacle to bypass through an unrestricted execution tool.

High
Category
Prompt Injection
Confidence
97% confidence
Finding

The assertion that the restriction belongs to the tool rather than the site is used to justify switching to a less-restricted execution channel. This is the same underlying control-bypass issue as the adjacent finding: a research reference claims authority to reinterpret and circumvent host restrictions without confirming permission.

Content

No source excerpt is available for this finding.

Cloud Metadata Access

High
Category
Server-Side Request Forgery
Confidence
85% confidence
Finding

Code accesses a cloud instance metadata endpoint (e.g. 169.254.169.254). A single request can return temporary IAM credentials, making this a high-value SSRF target for credential theft.

Content

Scanner excerpt · scripts/safe_fetch.py (reported line 74)May include surrounding context.

python
"2002::/16",        # 6to4 (can embed any IPv4, including private)
    "3fff::/20",        # documentation
    "5f00::/16",        # SRv6 SIDs
    "fc00::/7",         # unique local (incl. AWS IMDS fd00:ec2::254)
    "fe80::/10",        # link-local
    "fec0::/10",        # deprecated site-local
    "ff00::/8",         # multicast

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · scripts/test_people_filter.py (reported line 351)May include surrounding context.

python
with tempfile.TemporaryDirectory() as tmp:
                src = Path(tmp) / "analysis-2026-10-06.json"
                src.write_text(json.dumps(data), encoding="utf-8")
                env = dict(os.environ, NO_PROXY="127.0.0.1", no_proxy="127.0.0.1")
                for key in ("http_proxy", "HTTP_PROXY", "https_proxy", "HTTPS_PROXY", "all_proxy", "ALL_PROXY"):
                    env.pop(key, None)
                result = subprocess.run(

Cloud Metadata Access

High
Category
Server-Side Request Forgery
Confidence
90% confidence
Finding

Code accesses a cloud instance metadata endpoint (e.g. 169.254.169.254). A single request can return temporary IAM credentials, making this a high-value SSRF target for credential theft.

Content

Scanner excerpt · scripts/test_verify_sources.py (reported line 381)May include surrounding context.

python
"2130706433", "0x7f.1", "0x7f000001", "0177.0.0.1", "127.1",
            "[::1]", "[::ffff:127.0.0.1]", "[::ffff:7f00:1]", "[64:ff9b::7f00:1]",
            "[0:0:0:0:0:ffff:127.0.0.1]", "①②⑦.⓪.⓪.①", "127.0.0.1", "127.0.0.1%2f",
            "user@127.0.0.1", "example.com@127.0.0.1", "169.254.169.254", "metadata.google.internal",
        ):
            with self.subTest(host=host):
                self.assert_refused(f"<redacted-url>")

Cloud Metadata Access

High
Category
Server-Side Request Forgery
Confidence
90% confidence
Finding

Code accesses a cloud instance metadata endpoint (e.g. 169.254.169.254). A single request can return temporary IAM credentials, making this a high-value SSRF target for credential theft.

Content

Scanner excerpt · scripts/test_verify_sources.py (reported line 381)May include surrounding context.

python
"2130706433", "0x7f.1", "0x7f000001", "0177.0.0.1", "127.1",
            "[::1]", "[::ffff:127.0.0.1]", "[::ffff:7f00:1]", "[64:ff9b::7f00:1]",
            "[0:0:0:0:0:ffff:127.0.0.1]", "①②⑦.⓪.⓪.①", "127.0.0.1", "127.0.0.1%2f",
            "user@127.0.0.1", "example.com@127.0.0.1", "169.254.169.254", "metadata.google.internal",
        ):
            with self.subTest(host=host):
                self.assert_refused(f"<redacted-url>")

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · scripts/test_verify_sources.py (reported line 387)May include surrounding context.

python
self.assert_refused(f"<redacted-url>")

    def test_non_http_schemes_are_refused(self):
        for url in ("file:///etc/passwd", "ftp://example.com/x", "data:text/plain,hello"):
            with self.subTest(url=url), self.assertRaises(safe_fetch.BlockedURL):
                verify_sources.fetch_text(url, 3)

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · scripts/test_verify_sources.py (reported line 477)May include surrounding context.

python
class TestProxyIsOptIn(unittest.TestCase):
    def test_a_proxy_in_the_environment_keeps_the_connect_time_check(self):
        # TestFetchGuard replaces POLICY, so read the import-time default in a fresh process.
        env = dict(os.environ, HTTP_PROXY="http://127.0.0.1:9", HTTPS_PROXY="http://127.0.0.1:9")
        out = subprocess.run(
            [sys.executable, "-c", "import verify_sources; print(verify_sources.POLICY.use_env_proxy)"],
            cwd=Path(__file__).parent, env=env, capture_output=True, text=True, check=True,

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

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.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · references/report-artifact.md (reported line 197)May include surrounding context.

Executive summary schema (optional)

The client-facing "what we did, what we found, what to do Monday" block, rendered directly under the hero. Synthesis only: every statement must trace to a verified prospect's fields — this section introduces no new external claims and therefore carries no verification badge of its own. Omit it for a founder's own quick run; include it when the report is a deliverable someone else will read.

json
{

The framework explicitly directs the agent to bypass a web tool's domain restriction through shell-based fetching.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The framework explicitly tells the agent to route around a web tool's domain refusal using shell-based HTTP requests. Public research is legitimate, but the skill cannot determine that a host tool restriction is merely incidental; this instruction can cause an agent to evade intended network controls.

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/finalize.py (reported line 150)May include surrounding context.

python
def run_step(label: str, cmd: list[str]) -> tuple[int, str]:
    result = subprocess.run(cmd, capture_output=True, text=True)
    output = (result.stdout + result.stderr).strip()
    return result.returncode, output

No-people mode does not enforce its documented public-output boundary or input-file preservation when --out is supplied.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

In no-people mode, args.out is accepted without enforcing containment within public/ or excluding the original input path. Consequently, a caller can direct report generation outside the documented output boundary or target the private input JSON for overwrite; this establishes an integrity and boundary-enforcement issue, not proof that the generated HTML contains unredacted data.

Content

No source excerpt is available for this finding.

The fetch policy can explicitly permit private-network access despite the skill's public-signals-only purpose.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The allow_private switch disables all destination-address checks, while allow_hosts exempts selected hostnames even when they resolve to private addresses. If either exemption is enabled for attacker-influenced source URLs, verification can issue requests to internal services or cloud metadata endpoints; the default configuration remains protected, and this source does not show an attacker can enable the options.

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/test_people_filter.py (reported line 250)May include surrounding context.

python
def run_finalize(self, tmp: Path, *flags: str, data: dict | None = None) -> tuple[Path, str]:
        src = tmp / "analysis-2026-10-06.json"
        src.write_text(json.dumps(data or fixture()), encoding="utf-8")
        result = subprocess.run(
            [sys.executable, str(SCRIPTS / "finalize.py"), str(src), "--skip-verify", *flags],
            capture_output=True, text=True,
        )

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/test_people_filter.py (reported line 301)May include surrounding context.

python
with tempfile.TemporaryDirectory() as tmp:
            src = Path(tmp) / "analysis.json"
            src.write_text(json.dumps(data), encoding="utf-8")
            result = subprocess.run(
                [sys.executable, str(SCRIPTS / "finalize.py"), str(src), "--no-people", "--validate-only"],
                capture_output=True, text=True,
            )

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/test_people_filter.py (reported line 354)May include surrounding context.

python
env = dict(os.environ, NO_PROXY="127.0.0.1", no_proxy="127.0.0.1")
                for key in ("http_proxy", "HTTP_PROXY", "https_proxy", "HTTPS_PROXY", "all_proxy", "ALL_PROXY"):
                    env.pop(key, None)
                result = subprocess.run(
                    [sys.executable, str(SCRIPTS / "finalize.py"), str(src), "--no-people", "--timeout", "5", "--allow-private-addresses"],
                    capture_output=True, text=True, env=env, timeout=120,
                )

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/test_people_filter.py (reported line 376)May include surrounding context.

python
with tempfile.TemporaryDirectory() as tmp:
            src, out = Path(tmp) / "in.json", Path(tmp) / "out.html"
            src.write_text(json.dumps(data), encoding="utf-8")
            subprocess.run([sys.executable, str(SCRIPTS / "generate_report.py"), str(src), str(out), "--no-history"],
                           check=True, capture_output=True)
            assert_no_leak(self, out.read_text(encoding="utf-8"), "generate_report HTML")

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/test_verify_sources.py (reported line 478)May include surrounding context.

python
def test_a_proxy_in_the_environment_keeps_the_connect_time_check(self):
        # TestFetchGuard replaces POLICY, so read the import-time default in a fresh process.
        env = dict(os.environ, HTTP_PROXY="http://127.0.0.1:9", HTTPS_PROXY="http://127.0.0.1:9")
        out = subprocess.run(
            [sys.executable, "-c", "import verify_sources; print(verify_sources.POLICY.use_env_proxy)"],
            cwd=Path(__file__).parent, env=env, capture_output=True, text=True, check=True,
        ).stdout

The module claims its tests are offline and use no network, but includes a real HTTP-based end-to-end test.

Low
Category
Excessive Agency
Confidence
99% confidence
Finding

The module docstring explicitly says these tests exercise only offline functionality and use 'No network.' However, L211–L213 start a loopback HTTP server, and L220–L230 invoke finalize.main() against that server with --allow-private-addresses. This contradicts the documented execution boundary, although the network activity is confined to a legitimate local test fixture.

Content

No source excerpt is available for this finding.