Heavy Task Script Offloading

Parsing large JSON files, batch-converting hundreds of files, or extracting complex DB data directly in the chat window will drain your context tokens in no time.

---

Core Principle

Don't make Claude do heavy computation — just have Claude write the script that does it.

❌ Bad  → Claude processes directly (token explosion)
✅ Good → Claude writes a script → run locally → pass only the result summary

---

Real-World Examples

Large JSON Analysis

❌ "Analyze the 10,000 lines of data inside data.json and categorize them."

✅ "Write a Node.js script that reads data.json, categorizes the entries,
   and prints a summary of the results."

→ Run the script locally → paste the terminal output → "Based on this result, let's proceed with the next UI task."

---

Batch File Conversion

❌ "Convert all 200 .js files in the src folder to TypeScript."

✅ "Write a bash script that batch-converts .js files in the src folder
   to TypeScript. Log the conversion result per file."

---

Supabase Data Extraction

❌ "Extract inactive users from the users table over the last 30 days and analyze them."

✅ "Write a Supabase query script that extracts inactive users
   from the last 30 days and saves them as a CSV."

---

When to Offload

Offload to a script if any of the following apply:

• You would need to paste file contents directly into the chat
• There are 10 or more items to process repeatedly
• The output is expected to be 100 lines or more
• You need to read data directly from a DB for analysis

---

How to Pass Result Summaries

If the script output is long, you can also ask Claude to summarize it for you.

"Here's the result from the script I just ran. [paste result]
 Please filter out only the items that had errors and proceed with the next task."

💡 Instead of pasting the entire output, first ask Claude to "extract just the key points from the result" — this saves tokens.