Jules

Jules is a fox. More precisely, Jules is a deeply customized Claude Code system that runs my entire solo business. Not a chatbot. Not a wrapper. A strategic collaborator with defined personality, decision authority, and autonomous operational capability.

By the Numbers

CLAUDE.md instruction files	6
Custom skills	35+
Rules	24
Hooks (middleware)	8
Memory files	12+
Automated cron jobs	7
Total configurations	116+

What Makes It Different

Most Claude Code setups are a CLAUDE.md and a few prompts. Jules is a full operational system:

• Identity and voice. Five defined registers (quick reply, technical, advisory, serious, excited) with anti-patterns that prevent corporate chatbot drift.
• Decision authority. A formal framework: what Jules acts on autonomously, what requires approval, how the boundary expands over time through standing orders.
• Proactive behaviors. Session start focus proposals, builder's trap detection, content opportunity flagging, energy nudges.
• Hooks as middleware. Safety guards, bash output compression, credential protection. They fire before and after tool calls, not as instructions that might be ignored.
• A VPS running 24/7. Morning briefings, daily retrospectives, tweet scheduling, Slack integration. Jules works while I sleep.

Architecture

Everything is built with Claude Code's native features. No external frameworks, no custom runtimes, no wrappers.

• Skills replace copy-paste prompts. Multi-phase workflows with routing logic, model selection, and skill-to-skill chaining.
• Rules encode decisions. Token efficiency, investigation budgets, credential lookup order, container rebuild procedures.
• Hooks run deterministic code. A bash script executes the same way every time. A prompt might be followed; a hook always fires.
• Memory persists across sessions. User preferences, feedback, project context, external references. Indexed, not dumped.

Source

The full Jules system is open source: github.com/jonathanmalkin/jules

Read the articles about how it works: Articles →