Automation & Cloud Execution
From local CI/CD scripts to cloud-hosted agents that run while your laptop is closed.
The -p flag turns Claude Code into a composable Unix utility: feed it stdin, get structured output, and chain it with grep, jq, and awk. This article covers every output format, structured output with JSON Schema, resource limits, and real automation patterns.
Claude Code integrates cleanly into GitHub Actions and GitLab CI via the -p flag. This article covers permission design for headless environments, cost control strategies, three ready-to-use workflow patterns, and the --from-pr flag for PR-linked sessions.
Claude Code can drive a Chrome browser to navigate, click, fill forms, and take screenshots — enabling a self-verification loop where Claude implements a feature and then tests it in the running app without human intervention.
The /loop command runs a prompt on a recurring schedule — either at a fixed interval or self-paced where Claude decides when to check back. It's the foundation for CI monitoring, log watching, and the ticket-to-PR automation loop.
The /schedule command creates recurring remote agents that run on cron schedules, separate from your interactive session. This article covers the difference between scheduled agents, hooks, and /loop — and when to use each.
Channels stream Claude Code session events to external platforms like Telegram, Discord, and iMessage in real time. They're the outbound complement to MCP push notifications — giving you and your team live visibility into what Claude is doing.
For most GitHub workflows, gh CLI outperforms the GitHub MCP server: no context overhead, battle-tested, zero setup if you're already authenticated. This article covers when to use gh vs GitHub MCP, core patterns, and building automation workflows around gh.
Routines are saved Claude Code configurations that execute on Anthropic's cloud infrastructure, not your machine. Understanding when to use Routines versus /loop versus local scheduled agents is the foundation for building reliable automation.
A Routine is defined by three components: the prompt, the repository it operates on, and the connectors it can use. Getting each one right is what separates Routines that run reliably for months from Routines that stall, overshoot, or produce noise.
Routines consume two resources: run credits from your daily cap and tokens billed on top of that cap. Understanding how both work — and how to optimize for each — determines whether your Routines fit your budget or blow it.
Four high-value production patterns cover the majority of Routine use cases: deploy verification, nightly backlog triage, documentation drift detection, and dependency security audits. Each pattern has distinct trigger requirements, connector needs, and prompt design characteristics.
Creating a Routine is the beginning, not the end. Effective Routine management means knowing how to monitor, edit, debug failures, and make the right call between disabling and deleting. This article covers the full operational lifecycle.
Remote Control, Dispatch, and Channels are three distinct but composable features that transform Claude Code from a local terminal tool into a remotely observable and controllable service. Each solves a different problem — using the wrong one adds complexity without value.
Remote Control lets you connect to an active Claude Code session from any device, observe the full transcript in real time, and inject messages to steer the session. The session keeps running where it started — Remote Control is a transparent window into it.
Dispatch is a managed async task queue purpose-built for Claude Code. Unlike a generic message queue, it respects agent permission profiles, supports dependency chaining between jobs, and integrates natively with Channels for real-time status streaming.
Channels provides a continuous structured feed of everything Claude Code is doing — every tool call, response, error, and completion — streamed in near real-time to Telegram, Discord, iMessage, or a custom consumer. This is deep observability, not just completion notification.
Three composable patterns cover the majority of scenarios that genuinely require combining Remote Control, Dispatch, and Channels: the monitored long-run, the CI response pipeline, and the team agent queue. Each pattern combines specific primitives for specific reasons — not by default.