Set up dev environment + add AGENTS.md cloud notes

[devinoldenburg]

Summary

• Set up and verified the fm-bench development environment (Node 22, no runtime deps): npm install, npm test (42 pass), npm run lint (clean).
• Documented Cursor Cloud setup/run guidance in a new AGENTS.md, including the key gotcha that Apple's fm CLI is unavailable off macOS and how to drive the benchmark via the FM_BIN override.
• No application code changed.

Demonstrated the core benchmark end-to-end on Linux by pointing FM_BIN at a stub fm (kept outside the repo) that emulates the --help/available/quota-usage/token-count/respond subcommands. Verified doctor, models, the full benchmark, validate, export, compare, history, and the --ci SLO gate.

fm_bench_doctor_and_benchmark_demo.mp4

doctor then a 18-measurement system,pcc benchmark rendering result tables (100% success).

Benchmark result tables

Verification

• npm test
• npm run lint
• Real fm smoke test on macOS 27+, if benchmark behavior changed (N/A — no behavior changes; macOS fm unavailable on Linux CI, used FM_BIN stub instead)

_{To show artifacts inline, enable in settings.}