/ai-usage — unified AI usage report

Run the platform launcher from this skill's directory: ./run.sh on macOS or Linux; py -3 run.py on Windows, falling back to python run.py if py is unavailable. Reformat its output into the table below. Nothing else. No browser, no stored credentials, and no direct HTTP calls of your own.

The launcher starts all five providers and their local session reads concurrently. Benchmark requests start as soon as model/effort is known, overlapping the usage lookups. Results are printed in provider order without mixing output. Each provider lookup gets at most 20 seconds total, including CLI startup and every request in that probe. On timeout it stops the probe and its CLI children, prints a timeout for that provider, and continues without retrying. The benchmark lookup has the same 20-second budget; if it times out, report its failure and keep the usage blocks. Never extend a timeout for a missing subscription.

/ai-usage fresh — same thing; the script always reads live. The flag only means "do not reuse an answer from earlier in this conversation".