Cody skill library
Focused operating procedures for discovery, browser work, testing, debugging, and skill creation.
6
hand-picked skills
6
with 5K+ installs
3.2M
highest install count
6 skills
Open category page
vercel-labs/skills
Lets an agent search the open skills ecosystem when a task calls for specialized knowledge. Best for people who want a small on-demand toolkit instead of preinstalling a huge collection.
vercel-labs/agent-browser
Adds persistent, command-driven browser automation for navigation, extraction, form work, and web QA. Best for repeatable browser tasks and web application checks where a structured CLI is more reliable than ad hoc scripts.
obra/superpowers
Keeps an agent in a root-cause investigation loop before it starts patching symptoms. Best for intermittent failures, regressions, and multi-component bugs where plausible guesses are expensive.
mattpocock/skills
Guides vertical, behavior-focused red-green-refactor cycles instead of writing a large test suite up front. Best for features and bug fixes where a clear observable contract can drive implementation.
mattpocock/skills
A diagnosis loop for stubborn failures and performance regressions where the cause is not yet known. Best for hard bugs, slow paths, and ambiguous failures that have already resisted a straightforward fix.
anthropics/skills
A full workflow for drafting, testing, measuring, and improving an agent skill. Best for teams turning repeatable internal procedures into maintained, evaluable agent capabilities.
How the list was made
Cody's editors selected a focused set of agent skills that solve recognizable jobs, come from inspectable GitHub repositories, and show meaningful adoption or active maintenance.
Prefer focused skills that encode a repeatable workflow or a maintained domain playbook.
Favor clear ownership, active repositories, and meaningful adoption, then use community experience as a supporting signal.
Explain what each skill does, who it helps, and what to review before installing it.
Keep the collection small enough that users can install only what matches their work.
Browse the rebuilt prompt library for detailed, model-aware templates designed for GPT-5.6 and Claude Fable 5.1.
Explore prompt templates