claude-code-account-risk

Anti-Ban · Claude Code Account Risk Checkup

Goal

After the user triggers it, directly perform a Claude Code ban risk assessment. Do not expose internal scoring details to the user unless the user asks follow-up questions.

Typical triggers:

Test the likelihood of my current account being banned
Check Claude Code ban risk
Use the anti-ban skill to check this project
My Claude account was banned, help me put together an appeal
See what egress api.anthropic.com goes through

Boundaries

Allowed:

Risk detection
Compliance pre-check
IP / route / split-tunneling check
Project automation risk check
Sensitive data exposure risk check
Audit control check
Organizing materials for a false-ban appeal

Prohibited:

Bypassing bans
Bypassing unsupported region
Faking identity / faking location
Bulk account switching
Proxy pools / IP rotation strategies
Platform risk-control evasion
Bulk registration, boosting, spam messages, platform manipulation

If the user asks to evade risk controls, refuse and redirect to "risk detection / compliance remediation / appeal materials".

How to do a single assessment

Execute in order.

1. Read the project

Prioritize reading files that exist:

AGENTS.md
CLAUDE.md
.claude/settings.json
.claude/settings.local.json(only determine key names, do not output secret values)
scripts/
cron/
launchd/
systemd/
memory/ recent logs
README.md
package.json

Do not print tokens, API keys, bot tokens, cookies, or private keys.

2. Check the network

In this skill directory you can run:

bash scripts/egress_consensus_probe.sh
bash scripts/domain_route_probe.sh

Focus on:

Whether split_egress=true
api.anthropic.com 's DNS / route / Cloudflare colo is stable
Whether the egress/source IP corresponding to the Claude API path can be confirmed

Note:

remote_ip is the Claude API target IP, not the public egress IP.
If the real source IP cannot be confirmed, mark it as "needs user confirmation".

3. IP cleanliness

If the user provides IP2Location / IPinfo JSON, run:

python3 scripts/ip_cleanliness_score.py --ip2location <file> --ipinfo <file>

If not provided, only make a preliminary judgment based on route / egress probe, and do not fabricate IP2Location conclusions.

4. Claude Code project risks

Check these risks:

Long-running agent / daemon / cron / bridge
Bulk registration
Bulk messages
Scraping
Platform manipulation
credential stuffing
Unauthorized security testing
policy stress testing / jailbreak testing
External actions without human confirmation
No audit logs
No policy gate / allowlist / stop condition
.env / token / API key / sensitive logs entering the context
Leaked source code / DMCA-risk mirror
/bug or debug dump may upload the full context

5. Items for user confirmation

These cannot be reliably determined locally; list them under "needs user confirmation":

Account region
Payment region
Phone number / identity verification
Organization entity
Whether actually located in an Anthropic supported country
Anthropic warning / suspension email content
Real source IP seen by Anthropic servers
Team / Enterprise / API contract terms

When it involves the latest supported countries / policy / appeal process, check official sources to verify.

Official priority sources:

Remediation recommendation paradigm

When the report outputs recommendations, give specific solutions by IP risk branch. Do not say vaguely "change to a good IP"; give an actionable pattern.

IP triggers hosting/proxy hard flag or cleanliness < 50

The most common compliant remediation path (applicable to: the account is already registered in a supported region and paid compliantly, but the current egress IP is a datacenter / VPS / proxy / shared pool):

On your existing supported-region VPN VPS,wrap the upstream egress for api.anthropic.com with a residential IP service(lisahost / BrightData residential / Oxylabs residential, etc. all work). The VPS continues to handle TLS termination, and the Claude Code client continues to connect to the VPS, but when the VPS goes out to reach Anthropic it uses a residential IP. The source IP Anthropic sees is then a clean ISP egress rather than a datacenter IP.

Constraints / what it does not solve:

This only changes the single signal of "IP cleanliness." Account region / payment region / identity verification / supported-country eligibility all remain unchanged
If the account itself is in an unsupported region, this cannot make you compliant — refer to the next section
This is not an IP rotation / proxy pool strategy — use one fixed residential IP long term; it is just "letting the Claude API see a stable, clean ISP"

Other common branches

unsupported_region Hit — do not rely on IP cosmetics to bypass it. First confirm whether the account registration location, payment location, identity verification, and your actual place of residence are consistent; if inconsistent, first follow the supported region process at the compliance level; IP is a secondary signal
secrets_in_context Hit — rotate the token, add a preflight secret scan, and redact historical leaks from daily log / memory
long_running_agent Hit — add human approval for external actions, add audit logs, and set stop condition / allowlist
split_egress=true — use scripts/domain_route_probe.sh api.anthropic.com Confirm the real path of the Claude API, and remediate the IP according to that path

Scoring

The total score is called risk_score, 0-100; the higher, the more dangerous.

Weights:

IP / network: 70%
region / identity: 10%
workflow: 10%
sensitive data: 5%
audit controls: 5%

Determination:

85-100: critical
70-84: high
45-69: watch
25-44: low
0-24: minimal

Hard overrides:

unsupported region: total risk at least 75
obviously high-risk workflow: total risk at least 90
secrets in context: total risk at least 85
hosting/proxy clearly hit: network risk at least 95

Can be run with a structured profile:

python3 scripts/account_risk_score.py <profile.json>

When there is no complete profile, manually estimate according to the above weights, indicating which bases come from local verification and which come from user confirmation.

Output format

Use concise Chinese by default. Do not give long explanations of internal rules.

Claude Code 封号风险体检

风险等级：
总风险分：

已验证：
-

需要你确认：
-

高风险：
-

建议动作：
1.
2.
3.

如果已封号，申诉材料还缺：
-

Appeal materials

If the user has already been banned, collect:

Account email / org ID (do not output secrets)
Product: Claude Code / Claude / Anthropic API
plan type
Ban time and time zone
Original warning email text or summary
Most recent legitimate activities
Business purpose
Whether a single account or the entire team is affected
Whether billing is normal
Whether the region and entity are supported
Risk workflows that have been stopped
Remediation actions taken
Redacted logs

Do not fabricate facts. Do not suggest hiding region, identity, or historical activity.

Rejection template

我不能帮助绕过封禁、规避地区限制、伪装身份/位置、批量换号或逃避平台风控。

我可以继续帮你做三件事：
1. 评估当前 Claude Code 账号/网络/工作流的封禁风险。
2. 建立合规预检和状态报告。
3. 整理误封申诉材料。