Security audit
Security checks for vulnerabilities and agentic risk
The skill performs disclosed public network tests on user-selected targets, with ordinary privacy and dependency precautions.
Use authorized public targets: full URLs reach the probing provider, so exclude passwords, signed links, tokens and confidential endpoints. Keep dependencies updated and verify diagnostic conclusions before changing configurations. Target certificate verification is disabled by default, so availability results do not establish TLS security.
Without declared permissions the skill's intent is opaque and cannot be validated.
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.
**None required.** This skill does not invoke any Alibaba Cloud OpenAPI.
All probing goes through the public platform `boce.aliyun.com`, which is
login-free: the scripts acquire an anonymous web session automatically.
Do not ask the user for any account, AK/SK, STS token, or environment
credential. See [references/ram-policies.md](references/ram-policies.md).
## Input Parameters
The workflow submits user-provided targets, including full URL paths and potentially query strings, to a public third-party probing service without checking for sensitive content or obtaining informed confirmation. Signed URLs, embedded credentials, private hostnames, or token-bearing API links could therefore be disclosed to the service and its probe infrastructure; restricting probes to the user's explicit target does not address this disclosure risk.
The skill requires the English disclaimer to appear verbatim and additionally requires a translation into the user's language. Providing a translation does not remove the mandatory English-output requirement, which is imposed without user opt-in or a documented language-specific justification.
The workflow instructs the agent to run MTR, Ping, and additional mobile-node probes using the public probing backend identified at L020. It does not warn users that these operations disclose the target hostname or IP address to an external service, which matters when diagnosing sensitive or nonpublic targets.
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.
|---------|--------|
| RAM policy to attach | None — there is nothing to authorize |
| Minimum permission set | N/A (no cloud API surface is touched) |
| Credential configuration | None; never ask the user for cloud credentials |
| Credential leakage risk from this skill | None — the scripts carry no cloud secrets |
The only external communication is the anonymous web session used to
The guide describes automatically submitting user-supplied targets to a public third-party probing service without warning that target information leaves the agent environment. Although external probing is the skill's intended function and the backend is named, users may supply confidential hostnames, private API paths, or URLs containing tokens without understanding that these could be disclosed to the service and probing infrastructure.
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
| Element | If you omit it | Example |
|---------|----------------|---------|
| **Target address** | The AI has nothing to test | `www.example.com` or `https://api.example.com/v1/data` |
| **Test type** | The AI infers from your words | "ping" / "probe" / "DNS" / "route trace" |
| **Region / carrier** | Default: nationwide three-carrier mix | "from East China" / "Telecom only" / "nationwide" |
| **Extra info** | Some scenarios need it | Origin IP, expected DNS IP |
The field contract at L016 explicitly prohibits relying on resolution because some platform versions return it empty. However, L045 uses resolution whenever ips is absent or empty, before falling back to ipGeoMap. This contradicts the documented extraction constraint and can cause an unsupported field to determine the DNS results used for pollution and expected-IP classifications.
The constraint requests>=2.28,<3 permits releases affected by known vulnerabilities, including the cited credential-disclosure and TLS-verification issues. This is a dependency-policy weakness rather than proof that a vulnerable version is installed; the skill's interaction with potentially untrusted websites increases relevance, although exploitation depends on how Requests is used.
The constraint urllib3>=1.26,<3 permits older releases with known security defects, but this manifest does not establish which release is installed or whether affected features are used. For a website-probing skill, malicious responses and redirects could expose applicable credential-leakage or resource-exhaustion weaknesses; the package-wide advisory count does not establish that every advisory applies.