Best Practices
The difference between a good Workhog session and a great one comes down to how you provide context. These practices will make you 10x more effective.
Inspection vs. Raw Prompting
The single biggest productivity gain in Workhog is using inspection toolsinstead of describing your data in chat. Here's why:
1. Spreadsheet Data
"My CSV has a column called Amount with dollar values like $1,234.56 but some rows have just numbers without the $ sign. There are about 200 rows. Can you normalize them all to have the dollar sign and two decimals?"
~40 words of context
[Clicks cell B5] "Normalize this column"
Inspector auto-sends: column=Amount, pattern=currency, type=number, samples=[$1,234.56, 1234.56, $500, ...], 200 rows
3 words + automatic context
2. App Debugging
"My app isn't loading. I think it might be a data issue. The rows aren't showing up. I uploaded the CSV yesterday."
"Debug this app"
Debugger auto-captures: console logs, SDK state, file count, auth status, errors, screenshot of current view
3. Financial Analysis
"What's my revenue this month? The statements are in my workspace. Also show expenses and net profit. And what's the margin?"
"Summarize my financials"
The agent reads your uploaded statements and CSVs and answers with revenue, costs, margins, and top categories — every figure cited to its source file
The pattern
Notice the pattern: inspection tools provide structured context automatically. You write shorter messages, and Claude gives better answers. It's not about prompting harder — it's about giving Claude the right information with zero effort.
Prompt Engineering Tips
When you do need to write prompts (not everything has an inspector), here are tips that work specifically well with Workhog. For deeper dives into each inspector, see the CSV Inspector and App Tools guides.
Be specific about output format
Claude excels when you tell it what you want the output to look like:
Create a CSV summary of all invoices with columns: vendor, date, total, line_item_countSummarize my invoicesReference files by name
Don't say "my file" — say the file name. Claude searches your workspace by name:
Extract all line items from acme-invoice-q4.pdfAttach files for context
Click the + button next to the chat input to attach files directly to your message. This gives Claude the file content inline — no need to describe what's in it.
Chain operations in one message
Claude handles multi-step requests well:
Read payroll-2025.csv, calculate total annual cost per employee (salary + benefits), then create a new CSV sorted by total cost descendingUse the Prompt Supercharger
Workhog includes a built-in prompt refinement tool. Click the Supercharger icon in the toolbar to have AI help you write a better prompt. It's a multi-turn conversation that analyzes your draft and suggests improvements.
Model Selection Guide
Choosing the right model for the task saves time and gives better results:
| Task | Recommended Model | Why |
|---|---|---|
| Quick file search | Fast | Simple lookup, no deep analysis needed |
| Summarize a document | Balanced | Good balance of speed and comprehension |
| Analyze 50 invoices | Balanced | Handles moderate data well |
| Fill tax forms | Max | Complex multi-step with accuracy requirements |
| Fix a formula error | Fast | Quick targeted fix |
| Financial report | Max | Multi-source analysis, needs depth |
File Organization
A well-organized workspace makes the AI agent more effective. Here's a recommended structure:
documents/ — General documents, reports, memos
invoices/ — Invoice PDFs and CSVs
tax-forms/ — IRS forms, filled and blank
payroll/ — Payroll data and summaries
exports/ — Generated reports and outputsWorkflows & Saved Prompts
For tasks you repeat often, save prompts to your library:
- Write a prompt that works well
- Click the bookmark icon to save it
- Access saved prompts from the toolbar > Prompts panel
- One click to insert a saved prompt into any conversation
Good candidates for saved prompts:
- "Process all new invoices in the invoices/ folder and generate a summary CSV"
- "Generate a monthly P&L report from the statements in my workspace"
- "Check all CSV files for formatting inconsistencies"