Deep Dive Chapter 19: File Processing Workflows
You have 847 files in your Downloads folder. You need one of them by Friday. Three of them are duplicates eating 4 GB of space. And the file your boss asked about? It's in there somewhere.
You could spend an afternoon sorting manually. Or you could describe the problem in one sentence and let an agent handle it in 30 seconds.
But here's what this chapter is actually about: it's not about files. Files are the training ground. The real skill is learning to direct an AI agent: when to trust it, when to verify, when to demand a preview, and when to just do it yourself. Every pattern you learn here transfers to email management, project organization, data pipelines, and any domain where you work with General Agents.
By the end, you'll have a reusable prompt toolkit and (more importantly) the instinct to demand safety, verification, and scripts from every agent interaction. Most people prompt blindly and hope for the best. You're about to learn the systematic approach.
In Chapter 18, you learned the Four-Phase Workflow for Claude Code sessions: Explore → Plan → Implement → Commit. File processing expands this into a seven-step framework: Survey → Backup → Design Rules → Test → Execute → Verify → Document. The extra steps exist because file operations are irreversible: a moved file is moved, a renamed file is renamed. The Four-Phase Workflow assumed version control as your safety net. Here, your files don't have git revert, so backups, testing, and verification become explicit steps instead.
📚 Teaching Aid
What You'll Learn
By the end of this chapter, you'll be able to:
| Skill | What It Looks Like |
|---|---|
| Direct file surveys | "Help me understand what's eating my disk space" |
| Request safety-first operations | "Back up important files before making changes" |
| Design organization systems | "Create categories that make sense for my workflow" |
| Automate batch operations | "Rename these 100 screenshots with a consistent pattern" |
| Recover from mistakes | "Something went wrong: compare against my backup" |
| Search intelligently | "Find that PDF from 2023 about taxes" |
This isn't about learning bash. It's about learning to work effectively with General Agents.
Why This Matters
The patterns you learn in this chapter aren't just about files. They're the foundation for building AI Employees that automate these workflows entirely. Learn these patterns now or you'll struggle with automation later.
Every expert who works with General Agents mastered these fundamentals first. File organization is the perfect training ground because the problems are concrete, the feedback is immediate, and the patterns transfer everywhere.
This chapter also introduces the rhythm you will use for every agent workflow in this book: you define the goal and what correct looks like (10%), the agent does the work (80%), you check the result before accepting it (10%). That rhythm is Spec-Driven Development applied to file work: describe the outcome precisely, agent executes, you verify.
How This Chapter Teaches
Every lesson in this chapter follows the PRIMM-AI+ learning cycle: a five-stage structure that builds genuine understanding instead of copy-paste fluency. Here is how each stage applies to file processing:
| Stage | What You Do | In File Processing Terms |
|---|---|---|
| Predict | Write down what you expect before touching the agent | Name the commands you think the agent will run, estimate how many files match, note your confidence |
| Run | Direct the agent with a specific prompt | Give the agent a concrete instruction and watch it execute |
| Investigate | Write your own explanation first, then probe deeper | Explain in one sentence why the output looks the way it does, then ask the agent about edge cases |
| Modify | Change one variable, predict the new result, verify | Swap the folder, change the file extension, or adjust the date range and confirm the outcome shifts as expected |
| Make | Complete a similar task independently for a new folder | Apply the same workflow to a different location without using the lesson's example as a template |
Each stage is present in every lesson. The Predict stage is always AI-free: your notes come first, before any agent interaction. The Make stage is always the mastery gate: you pass when you complete the task independently, not when you follow the walkthrough.
Chapter Flow
| Lesson | Time | What You'll Do |
|---|---|---|
| 1. Your First Agent Workflow | 25 min | Survey your files through conversation |
| 2. The Safety-First Pattern | 20 min | Learn to require backups before changes |
| 3. The Organization Workflow | 25 min | Design and execute file categorization |
| 4. Batch Operations Workflow | 30 min | Transform repetitive tasks into scripts |
| 5. Error Recovery & Resilience | 20 min | Deliberately break things and recover safely |
| 6. Search & Discovery Workflow | 25 min | Find lost files through description |
| 7. Capstone: Your File Toolkit | 30 min | Build your personal prompt library |
Total time: Approximately 2 hours 50 minutes
Seven Principles Observed
You won't memorize these principles. You'll see them in action:
| Principle | You'll Observe The Agent... |
|---|---|
| P1: Bash is the Key | Using ls, find, grep, mv behind the scenes |
| P2: Code as Interface | Creating reusable scripts for repetitive tasks |
| P3: Verification | Checking results after each operation |
| P4: Decomposition | Testing on one file before batch operations |
| P5: Persisting State | Saving rules and logs to files |
| P6: Constraints and Safety | Creating backups before destructive changes |
| P7: Observability | Logging everything and showing progress |
Prerequisites
Before starting this chapter:
- Complete Part 1
- Have Claude Code installed and working
- Have a Bash-compatible terminal ready:
- macOS/Linux: Your built-in Terminal app works out of the box
- Windows: Install Git Bash for a quick start, or run
wsl --installin PowerShell for the full Linux experience via WSL - Stuck? Ask your AI agent: "Help me get a bash terminal working on my machine"
- Have Python installed for running scripts
- Have a folder with files to organize (Downloads recommended)
What You'll Build
By the end, you'll have:
| Deliverable | Purpose |
|---|---|
| Prompt toolkit | Reusable prompts for common file tasks |
| Organization rules | Documented categorization logic |
| Search patterns | Templates for finding lost files |
| Automation scripts | Agent-generated code you can reuse |
From Manual to Automated
The patterns you learn here become the foundation for AI Employees:
| This Chapter (Manual) | Automated Workflow |
|---|---|
| Manual prompting | Automatic file watching |
| One-time organization | Continuous organization |
| "Show me the plan" | AI decides and acts |
You're building the vocabulary and verification instincts you'll need for autonomous agents. In Part 3, you will encode these same file-handling patterns directly into AI Employees that run them autonomously, without a human in the loop for each step.