Claude Shield

Local-first privacy and proxy-consistency audit for Codex, Claude Code, and Agent Skills hosts.

简体中文 · Skill instructions · MIT License

Audit first. Change only what you approve.

Claude Shield checks observable network and browser contradictions without spoofing fingerprints, hiding automation, fabricating identity, or promising to bypass platform review.

Quick start

git clone https://github.com/AschoofAlpha/claude-shield.git "$HOME/.codex/skills/anti-claude-check"

Invoke $anti-claude-check in Codex. For Claude Code, install the same folder as ~/.claude/skills/anti-claude-check and invoke /anti-claude-check.

What you get

• A full read-only Windows collector for proxy, DNS, IPv6, browser, and documented Claude Code privacy settings, plus a limited POSIX environment summary.
• A bilingual Chrome/Edge page with a transparent 0–100 heuristic score, category bars, findings, LLM-ready text, redacted JSON, and a PNG report.
• Clash Verge/Mihomo checks for rule mode, system proxy, service/TUN state, strict-route, fake-IP, DNS hijacking, LAN access, and the actual policy selection when its local controller is available.
• Recommendations split into Must fix, Optional consistency, and Leave alone.
• Optional, reversible privacy environment-variable remediation. No automatic DNS, route, firewall, VPN, IPv6-adapter, device-ID, cache, or browser-fingerprint changes.

The score is local evidence, not a prediction of account approval or suspension.

Read-only collection

Windows PowerShell 7:

pwsh -NoProfile -File .\scripts\collect_windows_network.ps1

Windows PowerShell 5.1:

powershell.exe -NoProfile -ExecutionPolicy Bypass -File .\scripts\collect_windows_network.ps1

macOS or Linux:

bash ./scripts/collect_posix_network.sh

Collector output can contain local identifiers. Keep raw output local and let the Skill redact it before sharing.

Browser score and image report

Open assets/browser-audit.html in the normal Chrome profile, run the audit, then repeat in Edge. The page:

• reads ordinary browser-provided values without changing them;
• displays an English or Simplified Chinese score and findings;
• exports a CLI-compatible redacted JSON report;
• copies an ANTI_CLAUDE_BROWSER_REPORT_V1 block for an LLM;
• generates a localized PNG report.

The optional WebRTC test contacts Google's public STUN service once. Raw ICE addresses remain visible only on the local page; exported reports replace them with classifications.

Optional CLI

python -m pip install .
anti-claude-check audit --offline --format json
anti-claude-check browser import .\browser-audit.json

claude-shield remains as a compatibility command. The wheel includes the collector scripts, browser page, schema, Skill instructions, and bilingual README.

Privacy opt-outs

Preview first:

pwsh -NoProfile -File .\scripts\remediate_windows_network.ps1

After explicit approval, apply the documented Claude Code privacy environment variables:

pwsh -NoProfile -File .\scripts\remediate_windows_network.ps1 -Apply

The script writes a backup and prints the exact rollback command. The broad non-essential-traffic opt-out can disable optional Claude Code features, so enable it only when that tradeoff is intended. On POSIX, --apply creates a private environment file and prints the source command; it does not edit shell profiles or network settings.

Safety boundary

• No fingerprint spoofing, automation concealment, CAPTCHA bypass, or multi-account tooling.
• No fabricated identity, residence, billing, tax, or payment information.
• No automatic device-ID deletion, telemetry-cache deletion, or global network modifications.
• No claim that a configuration prevents account review or suspension.

Use current primary documentation for Claude Code privacy controls and rate limits. Treat third-party detector labels as opinions until live routing evidence corroborates them.

If this project finds a real leak or prevents an unnecessary destructive change, a GitHub star helps others discover it.