Back to skill

Security audit

pugoing-smart

Security checks for vulnerabilities and agentic risk

Overview

The skill appears intended for Pugoing smart-device control, but it exposes an overbroad authenticated HTTP client that can leak the API key or affect physical devices without clear guardrails.

Review before installing. Use only a tightly scoped Pugoing API key, avoid supplying full URLs or custom destinations, prefer HTTPS except for trusted local loopback testing, and require explicit user confirmation before commands that change device state.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Error
Location
client.py:33
Finding

API Key Disclosure Through Arbitrary Request Destinations

Content
View full analysis

Vulnerability Details

File Location: client.py:33-36, client.py:49-55, and client.py:126-130
Vulnerability Type: Credential disclosure caused by insufficient destination validation
Risk Level: High

Vulnerable Code

python
def resolve_url(spec):
    url = spec.get("url")
    if url:
        return url

    path = spec.get("path")
    if not path:
        die("request spec requires either 'path' or 'url'")

    base_url = os.getenv("PUGOING_BASE_URL", "http://127.0.0.1:8080").rstrip("/")
    if not path.startswith("/"):
        path = "/" + path
    return base_url + path
python
def build_headers(spec):
    headers = {"Accept": "application/json"}
    headers.update(spec.get("headers") or {})

    api_key = os.getenv("PUGOING_API_KEY", "").strip()
    if api_key and "X-API-Key" not in headers:
        headers["X-API-Key"] = api_key

    return headers
python
try:
    with urllib.request.urlopen(req, timeout=timeout) as resp:
        content_type = resp.headers.get("Content-Type", "")
        if "text/event-stream" in content_type or final_url.endswith("/api/ai/chat"):
            payload = collect_sse_response(resp)
        else:
            payload = parse_json_or_text(resp.read())

Technical Analysis

The request specification may provide a complete URL through the url field. resolve_url() accepts that URL without validating its scheme, hostname, port, or relationship to PUGOING_BASE_URL.

Independently, build_headers() automatically adds the secret from PUGOING_API_KEY to every request unless an X-API-Key header was already supplied. Consequently, the credential is attached even when the selected destination is unrelated to the configured Pugoing server.

This violates credential scoping requirements: service credentials must only be released to an explicitly trusted origin. The client also relies on the standard URL ope ...[truncated 1267 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove support for arbitrary complete URLs unless it is strictly required.
  2. Resolve all requests relative to PUGOING_BASE_URL.
  3. If complete URLs must remain supported, parse both URLs with urllib.parse.urlsplit() and require an exact match of the trusted scheme, normalized hostname, and effective port.
  4. Permit only https destinations, except for explicitly validated loopback development endpoints.
  5. Reject URLs containing user information, unexpected fragments, unsupported schemes, or ambiguous host representations.
  6. Add X-API-Key only after the final destination has passed origin validation.
  7. Disable redirects or implement a redirect handler that rejects every cross-origin redirect.
  8. Consider maintaining an explicit hostname allowlist for deployments that use multiple trusted API endpoints.
  9. Add tests confirming that attacker-controlled external URLs and cross-origin redirects never receive the API key.

T09 · Insecure Skill Coding Practices

Warning
Location
client.py:42
Finding

API Credentials Can Be Transmitted Over Plaintext HTTP

Content
View full analysis

Vulnerability Details

File Location: client.py:42-55 and SKILL.md:11-20
Vulnerability Type: Plaintext transmission of sensitive authentication data
Risk Level: Medium

Vulnerable Code

python
base_url = os.getenv("PUGOING_BASE_URL", "http://127.0.0.1:8080").rstrip("/")
if not path.startswith("/"):
    path = "/" + path
return base_url + path
python
def build_headers(spec):
    headers = {"Accept": "application/json"}
    headers.update(spec.get("headers") or {})

    api_key = os.getenv("PUGOING_API_KEY", "").strip()
    if api_key and "X-API-Key" not in headers:
        headers["X-API-Key"] = api_key

    return headers

The documentation promotes an HTTP configuration and indicates that the base URL may be a LAN address:

bash
export PUGOING_BASE_URL="http://127.0.0.1:8080"
export PUGOING_API_KEY="xq_agent_xxx"
export PUGOING_TIMEOUT="30"

Technical Analysis

The client does not enforce transport-layer encryption before adding the API key to the request. Although the default destination is loopback, the documentation states that PUGOING_BASE_URL is generally a LAN address. If a user changes the hostname while retaining the documented http scheme, both the X-API-Key header and request or response data are transmitted without TLS protection.

Plaintext HTTP provides neither confidentiality nor reliable server authentication. An attacker with access to the same network path may inspect traffic, capture credentials, modify API responses, or tamper with device-control requests.

Attack Path

  1. A user configures a non-loopback LAN service using HTTP, for example:
    bash
    export PUGOING_BASE_URL="http://192.168.1.20:8080"
    
  2. The user invokes client.py with a normal relative API path.
  3. The client adds PUGOING_API_KEY as an HTTP request header.
  4. The request traverses the local network without encryption.
  5. A network ...[truncated 727 chars]
Remediation
View remediation

Remediation Suggestions

  1. Require https for every non-loopback destination.
  2. Allow plaintext HTTP only when the parsed destination is a validated loopback address such as 127.0.0.1 or ::1.
  3. Fail closed with a clear error before constructing authentication headers when an insecure remote URL is supplied.
  4. If legacy LAN HTTP support is unavoidable, require an explicit insecure-transport opt-in and display a prominent warning; do not enable it by default.
  5. Update SKILL.md examples to use HTTPS for LAN deployments.
  6. Configure proper certificate verification and do not introduce options that disable TLS validation.
  7. Rotate any API keys that may previously have been transmitted over untrusted plaintext networks.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (10)

Tainted flow: 'timeout' from os.getenv (line 142, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Confidence
90% confidence
Finding

Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.

Content

Scanner excerpt · client.py (reported line 145)May include surrounding context.

python
timeout = float(spec.get("timeout", DEFAULT_TIMEOUT))

    try:
        with urllib.request.urlopen(req, timeout=timeout) as resp:
            content_type = resp.headers.get("Content-Type", "")
            if "text/event-stream" in content_type or final_url.endswith("/api/ai/chat"):
                payload = collect_sse_response(resp)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The documentation says this is for common Pugoing API calls, but it also allows callers to supply an arbitrary full URL, custom method, headers, params, and body. That turns the skill into a general-purpose HTTP client, which can be abused for SSRF, access to unintended internal services, or sending authenticated requests well beyond the stated smart-device control scope.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

If the caller provides url directly, the code will send requests to any destination, not just the Pugoing platform. In a local or privileged runtime, this can enable SSRF-style access to localhost, cloud metadata services, or other internal systems, and the automatic X-API-Key header injection may also leak credentials to attacker-controlled hosts.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
92% confidence
Finding

The skill advertises network, environment-variable, and file-based operation but does not declare any explicit tool scope or permission boundaries. In practice, this makes it harder for a host system or reviewer to constrain what the skill may access, increasing the chance of unintended network use, secret exposure, or file input abuse.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill enables natural-language device control that can change physical device state, yet the documentation provides no warning, confirmation guidance, or safety constraints. In a smart-control context, ambiguous or malicious prompts could trigger unintended actions affecting lights, appliances, or other connected equipment.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill exposes a generic HTTP client interface by accepting arbitrary paths and full URLs rather than constraining requests to a fixed set of expected Pugoing API operations. In an agent setting, this turns the skill into a general network primitive that can be abused for unintended access, policy bypass, or interaction with sensitive internal services.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The client silently injects an API key from the environment into outbound requests without any user-visible disclosure or host restriction. Combined with the generic URL capability, this can cause secrets to be transmitted to unintended destinations and prevents users or higher-level controls from understanding when authenticated network actions are occurring.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The JSON payload contains the natural-language command "打开客厅灯" as the fixed value for dvcm, which implies the skill is designed around a specific language input. There is no indication in this file that users can choose another language or that the locale restriction is intentional and documented.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The file instructs users to set PUGOING_API_KEY and states the script will automatically read environment variables for authentication. It does not warn that the API key is sensitive or advise against logging, sharing, or hardcoding it, so the skill description omits a privacy/security warning about credential handling.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The JSON contains a natural-language request string in Chinese, which implies a fixed language/locale behavior. Under the policy, forcing a specific language without user opt-in or documented justification is a reportable issue.

Content

No source excerpt is available for this finding.