Features section design for Claude Code
A features section answers "what does this do for me", not "list every button".
To get a features section like these in Claude Code: pick one below, open its Skill, copy it into Claude Code, Codex or Cursor and give it three to six features with a sentence each. The agent lays them out with the same cards, figures and motion.
The default an agent reaches for is three columns of icons and text, which makes every product look alike. The sections below work differently: some give each feature its own working demo (a terminal that runs a deploy by itself, a latency chart that draws itself), some count their spec figures up as they scroll into view, and some give each feature a whole screen.
When you write the features, give each one a single point and describe the result in the reader's words, not yours. Use real figures: the Skills tell the agent not to invent numbers, so leave a gap rather than a guess.
From real finished sites
CORTIVA AI infrastructure platform site · Paid Skill
Light feature cards: a switchable cold-start comparison, a world map with latency chips, a terminal that runs a deploy by itself and a chart that draws itself.
WORLDVIEW SAAS AI SaaS product landing page · Free Skill
The 3D Earth moves to one side while frosted-glass feature rows and a four-card solutions grid fill the other.
TORQEN Electric motorbike brand site · Paid Skill
One feature per screen over looping footage, with spec figures that count up from zero as they arrive.
OFFSET STUDIO Design studio portfolio · Free Skill
Service cards stacked on a dark ground, turning up one at a time as you scroll.
BIOLUME Biotechnology company site · Paid Skill
Two large cards, each pairing a micrograph with a laboratory photograph, explain the platform and the pipeline.
FAQ
Three to six. Fewer than three will not carry a section, and past six nobody reads to the end. If you have more, keep the six that matter most here and put the rest on a product page.
No. None of the sections above lean on a row of icons; they carry each feature with a working demo, a real figure or a photograph, which is much of why they do not look like the agent's default.