Use Cases
Youcan'tmakeLLMsdeterministic.
Butyoucancontroltheirharness.
Generative AI is inherently stochastic. xcaffold doesn't pretend to fix that—instead, it compiles strict, verifiable boundary environments so your agents behave predictably across any local or cloud IDE.
Switch Providers Without Rewriting
Evaluating Cursor after a year on Claude Code? Your .xcaf manifests compile to both. Switch targets with one flag — no config rewrite.
$ xcaffold apply --target cursor$ xcaffold apply --target geminiStart from What You Have
Already have a .claude/ directory with carefully tuned agents? xcaffold import converts it into portable .xcaf manifests in seconds. No starting over.
Audit Trail
Every compiled config is traceable to a .xcaf source file. xcaffold status detects unauthorized edits. Git history tells you who changed what and when.
project.xcaf.state MATCHTeam Consistency
Every engineer on the team gets the same agent setup. No more “works on my machine” for AI tool configs. Define once in .xcaf, compile everywhere.
xcaffold applyCompose Role-Scoped Environments
Define a blueprint that selects exactly the agents, skills, and rules for a specific role or workflow. xcaffold apply --blueprint backend-api compiles only what's needed.
$ xcaffold apply --blueprint backend-apiSee Your Resource Dependencies
Visualize how agents, skills, rules, and contexts relate to each other. xcaffold graph produces a dependency map you can reason about.
$ xcaffold graphOnboard Engineers to AI Tools Faster
New joiners get a working agent setup on day one — not a wiki page and a prayer. Run xcaffold apply in the repo and every coding assistant is configured identically to the rest of the team.
day one setupStop fighting the context window.
Deploy your first declarative agent harness today to standardize the chaos.