Claude Routines: The Hidden Costs Nobody Talks About

하이라이트

• Anthropic Routines allow automated prompt execution on Anthropic cloud infrastructure using custom connectors, environments, and schedules.

• The /schedule command in Claude Code CLI creates daily automated triggers for tasks like fetching RSS feeds and posting summaries to Slack.

• Pro plan users are restricted to a daily cap of five routine runs per 24-hour period.

• Remote routines bypass local security restrictions by using the web fetch tool through Anthropic infrastructure instead of the local bash curl tool.

• Automated GitHub event triggers require the Claude desktop app as the CLI currently only supports time-based schedules.

• Cloud-based routines require a dedicated repository containing a .claude folder to access custom skills, settings, or hooks.

타임라인

1. 00:00 - 01:23

   Automated Scheduling and Remote Infrastructure

   Anthropic continues to expand its cloud-based features by integrating managed agents and remote scheduling. This specific update allows tasks to run autonomously on a set schedule or in response to specific events like creating a pull request. The infrastructure aims to handle complex workflows without requiring the user to keep a local terminal or session active.

   • Routines execute prompts on Anthropic cloud infrastructure using custom connectors and environment triggers.
   • The system supports three primary execution methods: in-session loops, local/remote schedules, and event-based triggers.
   • GitHub event triggers and API post calls enable automation based on external developer workflows.
2. 01:24 - 04:40

   Workflow Configuration via Claude Code CLI

   Creating a routine to fetch newsletter data involves defining specific RSS URLs and a Slack connector for delivery. Using the web fetch tool is often more reliable than bash curl because it routes through Anthropic's secured infrastructure. Users must ensure the prompt is written to operate autonomously to avoid execution failures during permission prompts.

   • The /schedule command initiates a setup wizard for creating automated tasks directly from the terminal.
   • Network security layers prevent the bash tool from making outbound requests unless specific domains are whitelisted in a custom environment.
   • Connectors and permissions must be pre-configured to ensure routines run hands-off without requiring manual approval.
3. 04:41 - 05:41

   Pricing Structure and Usage Limits

   While routines draw from general subscription usage limits, they carry an additional cap to manage server resources. This daily cap serves as a mechanism for Anthropic to gauge infrastructure load during the preview period. The 20x Max plan increases this limit to 15 runs per day, offering more headroom for high-frequency tasks.

   • Routines are available exclusively for Pro, Max, Team, and Enterprise subscribers during the research preview.
   • Standard Pro subscriptions are limited to five routine runs every 24 hours.
   • Manual test runs during the configuration phase do not count against the daily routine limit.
4. 05:42 - 08:30

   Advanced Integration with GitHub and Custom Skills

   For tasks like automated PR reviews, the routine clones the specified repository to access code and custom hooks. This setup allows for complex logic, such as using an MCP tool for GitHub interactions, even when the user is offline. Successful execution depends on correctly mapping tokens and ensuring the cloud instance can replicate the local environment via the repository.

   • GitHub event triggers such as pull request reviews must be configured through the desktop app rather than the CLI.
   • Remote instances of Claude Code require a repository-based hook to load local settings or custom skills.
   • A .claude folder within the linked repository acts as the storage for custom skill logic and JSON settings.
5. 08:31 - 09:59

   Cost Analysis and Future Outlook

   The current pricing model makes Routines expensive for repetitive, small-scale tasks compared to using cheaper models like GLM 4.0 or GPT-4o via third-party webhooks. However, the ease of creating multiple routines through a single prompt provides a significant speed advantage for technical users. These updates suggest a shift toward more integrated, autonomous cloud features from Anthropic.

   • Alternative tools like Hermes Agent or Multica remain more cost-effective for high-volume automations.
   • Routines excel at rapid setup compared to the lengthy manual configuration required by traditional LLM agents.
   • Anthropic is positioning these features as building blocks for a more comprehensive cloud-based agent ecosystem.