Claude's Managed Agents get per-agent effort levels and 500 skills
By Steven Van ·
Anthropic also added webhooks for environments and memory stores, event-seeded sessions, and sub-agent event streaming.
Anthropic pushed a batch of features to Claude Managed Agents that read like a wishlist for anyone building serious agentic systems. You can now configure effort levels per agent, seed sessions with events, add up to 500 skills per session, use webhooks for environments and memory stores, and stream events for sub-agents.
Per-agent effort is the sleeper feature
Configuring effort levels per agent matters more than it sounds. In a multi-agent system, not every agent needs to think equally hard, a triage agent can run cheap and fast while a deep-reasoning agent gets more compute. Setting effort per agent lets you tune the cost/capability trade-off across a whole system rather than one global setting, which is how you keep a complex agent workflow both smart where it counts and affordable everywhere else.
500 skills changes what an agent can be
Raising the limit to up to 500 skills across all agents in a session is a big jump in capability surface. Skills are how an agent acquires task-specific competence, so 500 of them per session means an agent (or fleet of them) can carry a genuinely broad toolkit at once, closer to a capable generalist worker than a single-purpose bot. Combined with the broader skills-as-standard movement across the ecosystem, it signals how much weight Anthropic is putting on skills as the primary way to extend agents.
Webhooks and streaming for real infrastructure
The rest are the plumbing serious builders need: webhooks for environments and memory stores (so external systems can react to and feed agent state), event-seeded sessions (start an agent with context rather than a cold prompt), and event streaming for sub-agents (observe what nested agents are doing in real time). Together they make Managed Agents feel less like a demo API and more like infrastructure you'd run a product on, observable, integrable, and stateful.
Why it matters
Managed Agents is Anthropic's bet on hosted, production-grade agents, and this update is squarely aimed at teams building on it. For developers, the takeaway is that the platform is maturing fast toward the controls real deployments require: cost tuning, broad capability, observability, and integration. If you evaluated it early and found it thin, this is the round that fills in the operational gaps.
Sources: Claude Developers on X, Claude.