Back to skill

Security audit

promostandards

Security checks for vulnerabilities and agentic risk

Overview

The skill matches its PromoStandards purpose, but needs Review because it can send supplier credentials, business data, and purchase orders to externally configured SOAP endpoints without enforcing safe destinations or HTTPS.

Install only in environments where callers and supplier configuration are trusted. Prefer bundled registry or administrator-reviewed configs, block inline configJson from untrusted users, bind least-privilege supplier credentials, reject or override plaintext HTTP endpoints, and use preview-po/test endpoints before any production order submission.

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
scripts/config.py:370
Finding

Arbitrary SOAP Endpoints Enable SSRF and Credential Disclosure

Content
View full analysis

Vulnerability Details

File Location: scripts/config.py:370-391
Related Locations: scripts/ps.py:45-49, scripts/adapters/base.py:78-84, scripts/soap.py:70-73
Vulnerability Type: Server-Side Request Forgery and credential exfiltration
Risk Level: High

Vulnerable Code

python
# scripts/ps.py:45-49
if payload.get("configJson"):
    config = SupplierConfig.from_dict(payload["configJson"])
elif payload.get("config"):
    config = SupplierConfig.load(payload["config"])
python
# scripts/config.py:370-391
url = entry.get("url")
version = entry.get("wsVersion") or entry.get("version")
if not url:
    raise ConfigError(f"service '{code}' is missing 'url'")
if not version:
    raise ConfigError(f"service '{code}' is missing 'wsVersion'")
key = normalize_service(code)
overrides = entry.get("namespaces") or {}
if not isinstance(overrides, dict):
    raise ConfigError(
        f"service '{code}': 'namespaces' must be an object "
        "mapping prefix to URI"
    )
services[key] = ServiceEndpoint(
    service=key,
    url=str(url),
    ws_version=str(version),
    test_url=entry.get("testUrl") or entry.get("test_url"),
    status=entry.get("status"),
    namespaces={str(k): str(v) for k, v in overrides.items()},
)
python
# scripts/adapters/base.py:78-84
cred_id, password = self.config.require_credentials()
prefix = self.AUTH_PREFIX
return (
    soap.element(f"{prefix}:wsVersion", self.VERSION)
    + soap.element(f"{prefix}:id", cred_id)
    + soap.element(f"{prefix}:password", password)
)
python
# scripts/soap.py:70-73
req = urllib.request.Request(url, data=payload, headers=headers,
                             method="POST")
try:
    with urllib.request.urlopen(req, timeout=timeout) as resp:

Technical Analysis

The entry point accepts an inline configJson object or an explicit configuration file. ...[truncated 3271 chars]

Remediation
View remediation

Remediation Suggestions

  1. Separate trusted and untrusted configuration paths

    • Treat inline endpoint configuration and arbitrary config files as privileged administrative inputs.
    • For ordinary calls, permit only supplier names resolved through the bundled registry or signed, administrator-approved overrides.
    • Do not allow a request payload to select both credential references and arbitrary endpoint hosts.
  2. Enforce destination allowlisting

    • Bind each supplier identity to a set of approved production and test hosts.
    • Verify that every service URL matches the approved supplier-host mapping before resolving credentials.
    • Consider storing credential bindings outside caller-controlled configuration.
  3. Validate URLs before every request

    • Require the https scheme.
    • Reject embedded user information, malformed ports, fragments, and unsupported schemes.
    • Resolve the hostname and reject loopback, private, link-local, multicast, reserved, and unspecified IP ranges for both IPv4 and IPv6.
    • Protect against DNS rebinding by connecting to a validated address or revalidating immediately before connection.
  4. Control redirects

    • Disable redirects for SOAP POST operations unless they are required.
    • If redirects are supported, validate every redirect destination using the same scheme, hostname, and resolved-address policy.
    • Never forward credential-bearing SOAP bodies across origins automatically.
  5. Add regression tests

    • Verify rejection of http, file, loopback, RFC1918, IPv6 local, link-local, and metadata destinations.
    • Verify rejection of redirects to prohibited hosts.
    • Verify that credentials are not resolved or inserted until endpoint validation succeeds.

T09 · Insecure Skill Coding Practices

Error
Location
assets/promostandards_endpoints.json:1488
Finding

Credentials and Business Data Can Be Sent Over Plaintext HTTP

Content
View full analysis

Vulnerability Details

File Location: assets/promostandards_endpoints.json:1488-1506
Related Locations: scripts/provision.py:68-74, scripts/soap.py:70-73
Vulnerability Type: Cleartext transmission of sensitive information
Risk Level: High

Vulnerable Code and Data

json
{
  "service": "PRODUCT",
  "service_name": "Product Data",
  "version": "1.0.0",
  "url": "http://api.memoryinnovationpromo.com:8080/ImpProductDataService.svc",
  "test_url": "",
  "wsdl": "https://promostandards.org/wp-content/uploads/2025/07/ProductData-1-0-0-1-1.zip",
  "status": "Production"
},
{
  "service": "PPC",
  "service_name": "Product Pricing and Configuration",
  "version": "1.0.0",
  "url": "http://api.memoryinnovationpromo.com:8080/ImpPricingAndConfigurationService.svc",
  "test_url": "",
  "wsdl": "https://promostandards.org/wp-content/uploads/2025/07/PPC_1_0_0-1.zip",
  "status": "Production"
},
{
  "service": "INV",
  "service_name": "Inventory",
  "version": "2.0.0",
  "url": "http://api.memoryinnovationpromo.com:8080/ImpInventoryService.svc",
  "test_url": "",
  "wsdl": "https://promostandards.org/wp-content/uploads/2025/07/InventoryV2Final-1-1.zip",
  "status": "Production"
}
python
# scripts/provision.py:68-74
entry = {"url": capability.url, "wsVersion": capability.version}
if capability.test_url:
    entry["testUrl"] = capability.test_url
if capability.status:
    entry["status"] = capability.status
service_map[service] = entry
python
# scripts/soap.py:70-73
req = urllib.request.Request(url, data=payload, headers=headers,
                             method="POST")
try:
    with urllib.request.urlopen(req, timeout=timeout) as resp:

Technical Analysis

The bundled registry contains supported service endpoints using plaintext HTTP. build_config() copies the endpoint URL into runtime configuration unchanged. The SO ...[truncated 2396 chars]

Remediation
View remediation

Remediation Suggestions

  1. Require TLS for authenticated operations

    • Reject all http:// endpoints before credentials are resolved or a SOAP envelope is constructed.
    • Apply the same rule to registry-derived endpoints, bundled overrides, inline configuration, and explicit config files.
  2. Mark insecure registry entries as unsupported

    • During registry loading or provisioning, classify plaintext endpoints as insecure and omit them from drivable services.
    • Report a structured error explaining that an HTTPS endpoint or an administrator-approved secure tunnel is required.
  3. Validate TLS securely

    • Retain default certificate and hostname verification.
    • Do not add certificate-verification bypasses to support legacy suppliers.
    • Where higher assurance is needed, support administrator-managed CA bundles or certificate pinning rather than disabling validation.
  4. Constrain legacy exceptions

    • If business requirements mandate a legacy HTTP service, require an explicit administrator-only opt-in outside the ordinary action payload.
    • Route traffic through a mutually authenticated secure gateway or VPN.
    • Do not permit purchase orders, credentials, personal data, or other sensitive payloads over a direct plaintext connection.
  5. Improve validation and tests

    • Make provision.py --validate fail, rather than merely warn, for authenticated HTTP endpoints.
    • Add tests proving that registry-derived and custom HTTP endpoints cannot receive credentials.
    • Audit the complete bundled registry and supplier overrides for all remaining plaintext endpoints.
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
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (22)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The code's primary function is limited to discovering and reporting PromoStandards service metadata from the public endpoint registry. It fetches /services, /companies, and /companies/{code}/endpoints, normalizes registry record shapes, writes a JSON dump, and prints adoption/version summaries. This aligns only with the narrow part of the description about knowing which PromoStandards services a supplier supports. It does not perform the core declared client behaviors: calling supplier inventory/product/pricing/order SOAP services, handling credentials, onboarding via configuration for runtime service access, or normalizing cross-version business payloads. Therefore the supplied code chunk materially underdelivers relative to the declared purpose and has a different primary purpose: registry discovery rather than a full PromoStandards client.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared description presents a reusable, config-driven PromoStandards client skill whose purpose is to retrieve supplier inventory, product, pricing/configuration, ordering data, and supported services, including PO submission. The supplied code chunk is not that client implementation; it is a smoke-test script that uses such a client. Its primary function is operational validation: load/build config, verify credentials, call a fixed set of read-only endpoints, summarize results, and exit with status based on failures. This is materially different from the declared purpose. While some overlap exists—inventory/product/pricing-related reads and support detection—the code explicitly excludes PO submission, only probes endpoint reachability, and is focused on health checking rather than general supplier data access. Therefore the description does not accurately represent this code chunk.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
83% confidence
Finding

The skill declares access to secrets via environment variables and documents file/network-driven operations, but it does not constrain tool scope with explicit permissions or allowed-tools metadata. In an agent environment, missing scope declarations can permit broader-than-necessary access and increase the blast radius if the skill or surrounding orchestration is misused.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The manifest describes the skill as serving inventory, product data, pricing/configuration, purchase-order submission, and reporting which services a supplier supports. This file explicitly enumerates and operationalizes many other service types as supported registry data, indicating the skill’s actual service scope is broader than the description claims.

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 · assets/suppliers/cutterbuck.json (reported line 2)May include surrounding context.

json
{
  "_note": "Endpoints from the PromoStandards registry capture, except PRODUCT: the registry lists only Product Data 1.0.0, while Cutter & Buck's own integration document specifies 2.0.0 at ProductData200.asmx. The vendor document wins. If 2.0.0 misbehaves, fall back to https://api.cbcorporate.com/promostandards/ProductData.asmx at wsVersion 1.0.0. Cutter & Buck also publishes MED 1.1.0, ODRSTAT 1.0.0, OSN 1.0.0 and INVC 1.0.0, which this skill has no adapters for.",
  "supplier": "Cutter",
  "credentials": {
    "id": "${PS_CUTTER_ID}",

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 · assets/suppliers/cutterbuck.json (reported line 10)May include surrounding context.

json
{
  "_note": "Endpoints from the PromoStandards registry capture, except PRODUCT: the registry lists only Product Data 1.0.0, while Cutter & Buck's own integration document specifies 2.0.0 at ProductData200.asmx. The vendor document wins. If 2.0.0 misbehaves, fall back to https://api.cbcorporate.com/promostandards/ProductData.asmx at wsVersion 1.0.0. Cutter & Buck also publishes MED 1.1.0, ODRSTAT 1.0.0, OSN 1.0.0 and INVC 1.0.0, which this skill has no adapters for.",
  "supplier": "Cutter",
  "credentials": {
    "id": "${PS_CUTTER_ID}",

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 · assets/suppliers/cutterbuck.json (reported line 15)May include surrounding context.

json
{
  "_note": "Endpoints from the PromoStandards registry capture, except PRODUCT: the registry lists only Product Data 1.0.0, while Cutter & Buck's own integration document specifies 2.0.0 at ProductData200.asmx. The vendor document wins. If 2.0.0 misbehaves, fall back to https://api.cbcorporate.com/promostandards/ProductData.asmx at wsVersion 1.0.0. Cutter & Buck also publishes MED 1.1.0, ODRSTAT 1.0.0, OSN 1.0.0 and INVC 1.0.0, which this skill has no adapters for.",
  "supplier": "Cutter",
  "credentials": {
    "id": "${PS_CUTTER_ID}",

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 · assets/suppliers/cutterbuck.json (reported line 21)May include surrounding context.

json
{
  "_note": "Endpoints from the PromoStandards registry capture, except PRODUCT: the registry lists only Product Data 1.0.0, while Cutter & Buck's own integration document specifies 2.0.0 at ProductData200.asmx. The vendor document wins. If 2.0.0 misbehaves, fall back to https://api.cbcorporate.com/promostandards/ProductData.asmx at wsVersion 1.0.0. Cutter & Buck also publishes MED 1.1.0, ODRSTAT 1.0.0, OSN 1.0.0 and INVC 1.0.0, which this skill has no adapters for.",
  "supplier": "Cutter",
  "credentials": {
    "id": "${PS_CUTTER_ID}",

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 · assets/suppliers/cutterbuck.json (reported line 26)May include surrounding context.

json
{
  "_note": "Endpoints from the PromoStandards registry capture, except PRODUCT: the registry lists only Product Data 1.0.0, while Cutter & Buck's own integration document specifies 2.0.0 at ProductData200.asmx. The vendor document wins. If 2.0.0 misbehaves, fall back to https://api.cbcorporate.com/promostandards/ProductData.asmx at wsVersion 1.0.0. Cutter & Buck also publishes MED 1.1.0, ODRSTAT 1.0.0, OSN 1.0.0 and INVC 1.0.0, which this skill has no adapters for.",
  "supplier": "Cutter",
  "credentials": {
    "id": "${PS_CUTTER_ID}",

Unbounded Resource Access

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill allows unbounded resource consumption (API calls, storage, compute). Without rate limits or quotas, a compromised or misbehaving agent can cause denial-of-service or cost overruns.

Content

Scanner excerpt · scripts/_selftest.py (reported line 67)May include surrounding context.

python
self.raises = raises
        self.calls: list[dict] = []

    def post(self, url, data, headers, timeout=None):
        self.calls.append({"url": url, "body": data, "headers": headers})
        if self.raises:
            raise self.raises

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

This code asserts that product requests send localizationCountry>US<, which reflects a fixed locale expectation embedded in the skill behavior. The file does not indicate any user choice, opt-in, or documented justification for forcing a US localization, which matches the natural-language locale policy violation category.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The adapter hard-codes English as the default localization language in multiple public methods, which can force a specific language choice when callers do not explicitly provide one. This is a natural-language locale policy concern because the file does not offer opt-in language selection or document why English-only defaults are required.

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 · scripts/adapters/purchase_order_1_0_0.py (reported line 113)May include surrounding context.

python
# The request left the process; we cannot know whether the
            # supplier recorded it. This is the ambiguous case.
            raise EscalationRequired(
                f"sendPO for PO '{order_number}' failed with no confirmation "
                f"from {self.config.supplier}: {exc}. The order may or may "
                f"not have been received — do not resubmit without checking "
                f"with the supplier.",

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 · scripts/adapters/purchase_order_1_0_0.py (reported line 115)May include surrounding context.

python
raise EscalationRequired(
                f"sendPO for PO '{order_number}' failed with no confirmation "
                f"from {self.config.supplier}: {exc}. The order may or may "
                f"not have been received — do not resubmit without checking "
                f"with the supplier.",
                supplier=self.config.supplier,
                po_number=str(order_number),

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The code sets language=payload.get("language") or "EN", which forces English whenever the caller does not explicitly supply a language. Under the policy, imposing a specific language without opt-in is a natural-language policy violation unless the locale constraint is clearly documented and justified, which is not evident in this file.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

The adapter sends a SOAP request via self.post(...) after embedding self.auth_xml() and productID into the request body. While the code comments describe protocol/schema behavior, there is no user-facing warning, log, print, or explicit disclosure in this file that the skill transmits identifiers and authentication material over the network.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

get_filter_values constructs a SOAP body containing self.auth_xml() and productID, then submits it with self.post(...). The file does not include any visible disclosure that this operation contacts an external service and transmits request data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

This code file performs outbound network calls via self.post(...) after assembling request bodies containing product identifiers and localization fields. The file has comments/docstrings about request structure and parameter requirements, but no user-facing warning, logging, or disclosure that data will be sent to an external supplier service.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
94% confidence
Finding

The adapter hard-codes default locale values of country="US" and language="EN" in a public method signature, which is a natural-language locale policy choice applied unless the caller overrides it. The same pattern appears in other methods as well, and there is no visible user choice or documented opt-in in this file.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

This method defaults to country="US" and language="EN", forcing a specific locale unless the caller knows to override it. That constitutes a language/locale policy concern because the file does not indicate user choice or a documented regional limitation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The close-out method also applies country="US" and language="EN" by default, continuing the pattern of forcing a locale choice in code. No comments or surrounding documentation in this file indicate that this is user-selected or regionally required.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

The module docstring explicitly states that localizationCountry and localizationLanguage default to US/EN, and the methods repeat those defaults in their signatures. This imposes a specific locale/language choice unless callers override it, which is a natural-language locale policy concern under the stated rules.

Content

No source excerpt is available for this finding.