ClaudeFlag Skill

Install a Codex Skill that lets an agent run the same local diagnostic and summarize matched risk signals.

Install

Install the skill into ~/.codex/skills/claudeflag or $CODEX_HOME/skills/claudeflag.

curl -fsSL https://claudeflag.com/install-skill.sh | sh

Run

After install, ask Codex to use $claudeflag to diagnose the current machine.

Use $claudeflag to run the local diagnostic and summarize matched risk signals.

Output shape

The terminal report puts the score first, then separates reported Claude Code clues from local heuristics. Wide terminals use two columns; narrow terminals fall back to one column.

ClaudeFlag local diagnostic

  █████   █████
 ██   ██ ██   ██
  █████   █████
 ██   ██ ██   ██
  █████   █████
  High risk  88 / 100
  Result: You are definitely a "Claude China user".
  Scope: timezone and endpoint host are reported Claude Code clues; other rows are local heuristics.

What it checks

System timezone and UTC offset on the machine running the command.
LANG, LC_ALL, LANGUAGE and locale output.
Chinese font families through fc-list when available.
ANTHROPIC_BASE_URL host clues without reading API keys.
Proxy environment variables and git proxy settings, reduced to hostnames only.
China public DNS resolver clues from /etc/resolv.conf.
npm, pnpm, yarn and pip registry hostnames for China mirror clues.

Privacy model

The CLI and Skill use the same local shell diagnostic. They print results to stdout and do not upload anything.

Runs on the local machine or server only.
Does not read Claude account data, API keys, shell history or Claude config files.
No network request is made after the script has been downloaded.