Anthropic adds cron-scheduled agent runs and environment-variable secrets to Claude Managed Agents

Anthropic has shipped three developer-facing additions to its Claude Managed Agents platform alongside the June 9 Claude Fable 5 launch. The most significant is scheduled deployments: developers can now trigger agent sessions on a cron schedule directly from the Managed Agents API, without maintaining a separate scheduler. Two companion updates—environment variable credentials in vaults and a new session_thread_id field in webhook events—round out what amounts to a meaningful expansion of the platform's production readiness.

What's new

Scheduled deployments. As Anthropic's API release notes put it, "Claude Managed Agents now supports scheduled deployments, letting you run sessions on a cron schedule without managing your own scheduler." Developers define a schedule in the API; Managed Agents handles triggering, provisioning, and session lifecycle. This removes the need for external infrastructure—Lambda functions, systemd timers, or third-party schedulers—to drive periodic agent tasks. Sessions run with the full Managed Agents feature set, including tool use, MCP connections, and multi-agent orchestration.

Environment variable credentials in vaults. Anthropic's vault system, which stores secrets for secure injection into agent sandboxes, now accepts environment variable credentials alongside the previously supported OAuth and API-key types. Services that authenticate via environment variables—a common pattern for CLIs, language SDKs, and internal tooling—can now be configured once in the vault and injected automatically into every agent session.

session_thread_id in webhook events. The session.thread_* webhook events now include a session_thread_id field identifying the multi-agent thread that triggered the event. For teams running orchestrated agent networks, this field enables stable routing, logging, and cross-thread debugging.

Context

Claude Managed Agents launched in beta in early 2026 and has expanded rapidly: webhooks, multi-agent orchestration, self-hosted sandboxes, agent memory, and scheduled deployments have all shipped within a few months. The platform provides a fully managed execution environment—handling sandbox provisioning, tool execution, file I/O, and session state—with the goal of letting developers ship production agents without managing compute infrastructure.

The June 9 release lands on the same day as Claude Fable 5, Anthropic's most capable widely released model. Fable 5 introduces always-on adaptive thinking, 1M-token context, and a new refusal safety model; the Managed Agents features are API infrastructure updates that apply across models. The vault environment variable support is particularly timely: as agents take on more agentic workflows requiring access to external APIs, secrets management becomes a recurring friction point.

Why it matters

Scheduled deployments lower the operational cost of a common class of production agents: anything that needs to run on a recurring schedule—nightly data summaries, daily health checks, weekly report generation, continuous monitoring loops—previously required external infrastructure to trigger. That barrier is now gone for Managed Agents users.

Combined with secure environment variable injection, these features materially close the gap between a proof-of-concept agent and a production deployment. A developer can now write an agent that connects to internal services via environment-based credentials, runs every hour, and emits structured webhook events—all within Managed Agents, without additional infrastructure. For teams evaluating hosted agent platforms, these additions strengthen the case for staying on Anthropic's stack rather than building around it.