Chat basics
Prompt patterns that work well, and the ones that don't.
What the AI sees
When you send a message, HISAB attaches sheet-level context automatically: workbook name, sheet names, header rows of each sheet, and the size of each used range. The AI does not see cell values until it asks for them — at which point it calls a read_range tool and HISAB returns the data.
Effective prompts
Be specific about scope
- ✅ "In the Bank Statement sheet, find rows where Description contains 'TRF' and Amount > 10000."
- ❌ "Show me transfers."
Name the artefacts
- ✅ "Compare GL_Apr (sheet) against the trial balance in TB_Apr — flag any account where the rolled-up GL doesn't match the TB balance."
- ❌ "Reconcile these."
State the output you want
- ✅ "Output the mismatches in a new sheet 'Mismatches' with columns Account, GL, TB, Diff."
- ❌ "Show me what's wrong."
Slash commands
| Command | What it does |
|---|---|
/skills | List all available skills |
/skill <id> | Invoke a specific skill |
/sheets | Summarise every sheet in the workbook |
/range A1:D10 | Quick view of a range |
/undo | Revert the AI's last write |
/mode strict | Switch operating mode without leaving the chat |
/clear | Start a new conversation (keeps audit log) |
Multi-turn workflows
The AI remembers context within a conversation. Build up complex workflows in steps:
You: Pull the May trial balance from Zoho Books into a new sheet.
AI: [pulls 184 accounts to sheet "TB_May"]
You: Now reconcile it against the GL in this workbook.
AI: [reads GL sheet, computes per-account diff]
"Found 3 mismatches over $100..."
You: Draft a JE to close the variance accounts.
AI: [drafts JE, places in approvals queue]