One model, many ways to steer it.
Hooks, skills, subagents, MCP, agent teams, dynamic workflows, agent view, worktrees, conventions, policies, routines… it’s a lot of nouns. Pick one primitive and read it end to end — each card opens a full page with the definition, an analogy, the mechanism, who invokes it, when to reach for it (and when not), best practices, and setup. The hard badge means client-enforced; otherwise it’s soft, model-mediated. See the full badge legend →
Conventions
Soft, persistent guidance the model reads every session.
→ hardPolicies
Client-enforced governance that the model can’t override. Not the same as conventions.
→ hardHooks
Your code, run automatically on lifecycle events. Not the same as rules.
→Skills
Reusable instructions/procedures loaded on demand. A saved procedure you call “a routine” is this — not the new Routines feature (its own card below).
→MCP
A protocol that gives the model new tools & data from external systems.
→Subagents
Delegated workers in their own context; only a summary returns.
→ experimentalAgent teams
A lead + peer sessions that share a task list and message each other.
→ previewDynamic workflows
Claude writes an orchestration script that runs dozens-to-hundreds of agents.
→ previewWorktrees + Agent view
Git isolation per session, plus one dashboard to dispatch & watch many of them.
→ research previewRoutines
A saved config that runs itself on Anthropic’s cloud — on a schedule, a GitHub event, or an API call.
→/loop
Re-run a prompt on an interval inside an open session — cron syntax with a model deciding each tick. The in-session cousin of routines.
→/goal
Keep working turn after turn until a fresh small model confirms a condition is met. Condition-driven, where /loop is clock-driven.