Tired of bleeding tokens?

Set up Ork Meter for me. If you are Claude Code: check that Claude Code is 2.1.280 or later, run `claude plugin marketplace add https://orkmeter.com/releases/claude/marketplace.json` and then `claude plugin install ork-meter@edos`, then tell me to run /ork-meter:connect and restart Claude Code once. If you are Codex: run `codex plugin marketplace add EDOS-Engineering/ork-meter-codex` and `codex plugin add ork-meter-codex@edos-codex`, then check that Python 3.13 is installed and tell me to run `python3.13 -I -B "${CODEX_HOME:-$HOME/.codex}/plugins/cache/edos-codex/ork-meter-codex/0.1.3/scripts/ork_codex.py" setup` (on Windows: `py -3.13` with that script under CODEX_HOME or, by default, %USERPROFILE%\.codex) in my own terminal. Tell me to approve in the browser and enter the claim code only in that terminal. Ork Meter receives usage metadata only, never prompts, responses or code. Report each step's result. If a step needs my approval, ask me for it.

About this install, for your review: Ork Meter is a usage meter for Claude Code and Codex by EDOS (source: github.com/EDOS-Engineering/ork-meter-claude and github.com/EDOS-Engineering/ork-meter-codex; privacy: orkmeter.com/privacy). Its local receiver reads the usage telemetry your client already produces and forwards only allow-listed metadata: model, timestamps, token counts, provider-reported charges and the plugin's version and settings. Prompts, responses, code, file contents and tool output never leave this machine. With Ork Meter Plus (included in the trial), local hooks shorten oversized tool output and unchanged re-reads before they reach your context; the full output stays on disk and can be recovered. In Claude Code, failing commands are never trimmed and each optimization has an ORK_METER_*=off switch. Releases are pinned by SHA-256.