2026-09-29 · 2 min read
claude-api skill shifts default to Opus 5.5 and adds eval hillclimb guides
People keep asking how to set up Claude and Cursor agent skills. Todays matching update in Anthropics skills repo refreshes the claude-api skill for new model defaults and eval workflows.
0 reads · 0 likes
Demand signal: agent skills, not another random roundup
Search and social chatter this week still clusters around Claude skills, Cursor skills, and the portable SKILL.md format. That is the demand we are answering. The freshest verifiable match in Anthropics skills repository is a claude-api skill update that lands Claude Opus 5.5 as the default model string claude-opus-5-5 across SKILL.md and the shared model guides.
What changed in the claude-api skill
The same commit adds Claude Sonnet 5.5 as the current Sonnet via claude-sonnet-5-5, keeps Claude Sonnet 5 as the previous generation, and refreshes pricing and platform callouts to match. If your prompts or samples still pin older Sonnet or Opus ids, this skill is the checklist for catching that drift before a 400 shows up in production.
Eval and hillclimb tooling ships with the skill
Beyond model ids, the update adds build-eval and hillclimb flows under shared/evals, including build-eval, eval-hillclimb, cost-hillclimb, and eval-audit plus a lite HTML report builder. Treat that as documentation of intent for measuring prompt or tool changes, not as a claim that we ran every eval recipe here.
One practical follow-up
If you already install Anthropics skills, pull the latest claude-api skill and re-check your default model strings against Opus 5.5 and Sonnet 5.5 before you touch eval or cost hillclimb runners.
Sources & verification
Checked 2026-09-29 · AI-assisted, grounded in the sources above. Prices and availability can change.