IP地址归属、信誉、风控查询

IP Intelligence Fusion

Use this skill for a single public IPv4 or IPv6 address that the operator owns, that is publicly documented, or that the operator is explicitly authorized to investigate. The result is a time-stamped evidence report for human review. It is not an identity lookup, an automatic allow/deny decision, a location guarantee, or a platform-review tool.

Mandatory boundaries

• Accept exactly one public IP. Reject hostnames, private addresses, reserved addresses, batches, customer login logs, account logs, cookies, device identifiers, and device fingerprints.
• Treat an IP as potentially personal information when it is linked with a person, account, customer, employee, or login record. Do not infer that command-line possession proves authority.
• Supported work is limited to owned/public/authorized asset checks, public-information verification, authorized security operations, abnormal-request triage, and provider delivery acceptance testing.
• Do not assist with unauthorized person investigation, bulk personal-IP collection, location spoofing, account farming, bulk registration, platform-review evasion, CAPTCHA or login bypass, access-control bypass, port scanning, exploitation, attacks, or proxy forwarding.
• Do not log in, submit forms, handle CAPTCHA, defeat rate limits, or use a mirror when reading a provider's public page. Stop when a page requires access beyond ordinary read-only viewing.

Network consent workflow

1. Validate that the user supplied one public IP. If no IP is present, ask for it. Do not resolve a hostname on the user's behalf.
2. Start with local-only processing. The default CLI mode must not call a remote service. It may read a local evidence JSON file and render JSON, Markdown, or HTML.
3. Before any external query, explain the target IP, providers/domains, fields sent, and that the transfer may cross borders. Ask for explicit confirmation. This is an operational record, not proof of legal authorization.
4. Only after confirmation use --external. In an interactive terminal the CLI presents a YES prompt. In a non-interactive process pass --confirm-external as well. --self requires both flags; the IP discovery request is made only after confirmation.
5. Use the default fast profile unless the user explicitly requests --profile comprehensive. Do not silently expand the set of recipients.
6. If confirmation is absent or refused, produce a local report with not-requested source states where appropriate. Do not turn that state into error, zero risk, or a negative finding.

Examples:

Local-only JSON:
python scripts/ip_intelligence.py 8.8.8.8 --format json

Interactive external lookup:
python scripts/ip_intelligence.py 8.8.8.8 --external --profile fast

Non-interactive external lookup:
python scripts/ip_intelligence.py 8.8.8.8 --external --confirm-external --format json

The removed --include-raw option must not be suggested or accepted.

Provider and evidence rules

The CLI request layer permits only audited HTTPS hosts and rejects credentials in URLs, user information, non-standard ports, and unapproved redirect destinations. The current domains and collection methods are listed in references/providers.md.

IPinfo and AbuseIPDB credentials, when independently configured, are request headers. IPQualityScore, Scamalytics, and ipdata API adapters are disabled; those services can appear only as validated official public-page evidence. The old plaintext IP-API adapter is removed.

Public-page evidence must contain the exact target IP, an official HTTPS source URL, an observation time, and only the allowlisted normalized fields. Do not include raw responses, fn, email, abuse-contact, analysis, or personalized-hostname fields. Use the local evidence import only after the host has actually observed the official page.

Keep these distinctions in every report:

• success: validated structured evidence was returned;
• skipped: an enabled provider needs a missing configured credential;
• not-requested: external collection was disabled, or an API adapter is intentionally disabled;
• unavailable: an experimental source could not be read or parsed;
• error: an enabled source failed validation, transport, or upstream processing.

Absence of evidence is not low risk. Numeric risk comes only from upstream numeric scores. Boolean proxy, VPN, Tor, hosting, abuse, and bot signals remain contextual or unscored. Preserve provider identity, consensus, alternatives, conflicts, and timestamps.

Report handling

Generate the requested representation with the CLI. Reports may contain the complete IP, geographic region, organization/ISP, allocation or route prefixes, and network-risk labels. Set restrictive file permissions; do not place reports in public Issues, demo sites, public logs, or uncontrolled shared storage. Delete them under the operator's retention schedule.

The JSON report must include policy metadata and the data_policy declaration. In external mode, record external-confirmed and the provider domains that actually started a request. Reports must not contain upstream raw payloads, contact details, credentials, or API keys. Never call an IP safe based only on this report and never present it as proof of abuse.

Completion brief

Return a concise summary in the user's language containing the normalized target, risk score or unknown, confidence, contributing numeric sources, consensus facts, material conflicts, contextual signals, source-state counts, timestamp, and absolute paths to generated reports. State that the report is an aid for authorized human review, not a legal conclusion or an automatic platform decision.

If a network is unavailable, preserve explicit provider failures and continue with local evidence. If a public page is blocked or changes layout, keep it unavailable and report the gap. Do not invent values or bypass the restriction.

Compliance reminder

External requests may transmit an IP to a service provider outside China. The operator is responsible for checking authorization, notice, lawful basis, personal-information handling, data-export requirements, retention/deletion, and third-party terms. This skill and its MIT license are not legal advice and cannot establish compliance by themselves. Future support for customer or account logs requires a separate personal-information impact assessment and data-export design; it must not be added directly to v2.0.