Automated Claude Task Scheduler with Git Isolation

What It Is

Dreamer is an automation scheduler that runs Claude coding tasks without human supervision. Built as part of the Claude Matrix plugin available at https://claudeonrails.dev, it accepts scheduling instructions in either cron syntax or natural language like “every weekday at 9am” and executes prompts on a timer.

The implementation uses git worktrees to maintain isolation from active development branches. When a scheduled task triggers, Dreamer creates a separate worktree, spawns a new branch, executes the Claude prompt, commits any changes, pushes the results, and cleans up the temporary workspace. This approach prevents scheduled tasks from interfering with ongoing work while ensuring all modifications are properly tracked in version control.

Why It Matters

Autonomous code maintenance represents a shift from AI as an interactive assistant to AI as an independent team member. Development teams can now offload repetitive quality tasks that typically get deferred during active feature work - test maintenance, dependency audits, documentation updates - to run during off-hours.

The worktree isolation mechanism solves a practical problem with automated code changes: how to let AI modify a codebase without disrupting developers mid-task. By operating in separate branches, scheduled tasks produce reviewable commits rather than direct modifications to working directories.

This matters most for small teams where maintenance work competes with feature development. A three-person startup can schedule nightly documentation reviews or weekly dependency cleanup without allocating sprint capacity. The trade-off is accepting that some generated commits will need rejection or revision, but the time saved on successful runs often justifies occasional cleanup.

Getting Started

Dreamer ships with the Claude Matrix plugin at https://claudeonrails.dev. After installation, scheduling works through simple commands:

For more complex timing, cron syntax provides precise control:

Common patterns developers are running include daily test maintenance, where Claude identifies and patches failing tests each morning before the team starts work. Weekly dependency audits catch outdated packages and security issues. Nightly documentation passes ensure README files and inline comments stay synchronized with code changes.

The key to effective scheduling is writing prompts that produce reviewable commits. Instead of “improve the codebase,” specific instructions like “find functions longer than 50 lines and extract helper methods” generate focused changes that are easier to validate during code review.

Context

Traditional CI/CD systems run tests and checks but rarely modify code autonomously. GitHub Actions and GitLab CI can detect issues but typically require human intervention to fix them. Dreamer occupies a different space by actually generating fixes and improvements rather than just reporting problems.

The closest alternatives are tools like Dependabot for dependency updates or automated formatters like Black or Prettier. However, those handle narrow, well-defined tasks. Dreamer’s flexibility comes from Claude’s general-purpose capabilities - the same scheduler can handle test fixes, documentation updates, and code refactoring.

Limitations center on trust and verification. Autonomous code changes require robust review processes since AI-generated commits can introduce bugs or make questionable architectural decisions. Teams need clear policies about which scheduled tasks auto-merge versus requiring human approval. The worktree approach helps by making each scheduled run produce a discrete, reviewable unit of work.

Resource costs also matter. Running Claude prompts on schedule consumes API credits whether the changes prove useful or not. Teams should start with conservative schedules and expand based on the value-to-cost ratio of generated commits.