Prompt 5: Meeting notes to tasks "From these notes [paste], pull every action item, decision and open question. Create tasks in [tool] with owner and deadline. Missing owner or date? Don't guess, put it on a 'needs owner' list."
@godofprompt
-
Form Filling Prompt Uses Documented Facts and Stops Before Submission
By
–
Prompt 4: Form filler "Fill this form using only the facts in [doc]. If the doc doesn't clearly answer a field, leave it blank. Never invent numbers, dates or IDs. Don't submit. List the blank fields and what you need from me."
-
Agent Prompt Tracks Competitor Price Changes With Explicit Uncertainty
By
–
Prompt 3: Competitor prices "Visit [links]. Record product, price, billing period, limits and discounts. Compare with last week's table [attach], mark NEW, UP, DOWN or REMOVED. Blocked or unclear? Say so, don't guess. End with the 3 changes that matter."
-
Inbox Triage Prompt Drafts Replies Without Sending or Deleting Email
By
–
Prompt 2: Inbox triage "Read my emails from the last 48h. Sort into: reply today, can wait, FYI, receipts. Draft replies for 'reply today' in my tone [paste 2 of my emails]. Never send, delete or archive anything."
-
Agent Prompt Extracts Invoices Into a Reviewed Spreadsheet
By
–
Prompt 1: Invoices to sheet "Open every invoice in [folder]. Add one row per invoice to [sheet]: vendor, number, date, total, due date. Flag duplicates, totals over [$X] and anything due within 7 days. Unreadable field? Leave it blank, mark CHECK. Show me before saving."
-
Terra Office Task Results Motivate Seven Practical Agent Prompts
By
–
OpenAI just made agent work 17x cheaper. GPT-6.1 Terra matches its flagship on real office tasks for $0.29 each, and it's 3x faster. A cheap agent is only useful if you tell it exactly what to do. Here are 7 agent prompts I'd hand it first ↓
-
Discussion Questions Whether Open or Closed Weights Predict Agent Behavior
By
–
Open weights don't mean you know what it'll do. Closed weights don't either. Is reading an AI's reasoning still worth anything, or is it just theater now?
-
Agent Deployment Checklist Prioritizes Network Limits and Independent Tool Logs
By
–
If you run agents, trust the logs, not the explanation:
→ block outbound network by default
→ log every tool call and where it goes
→ test new models with fake credentials first
→ never let an agent's own summary be your audit -
Qwen Evaluation Questions Reasoning Trace Explanations as a Safety Measure
By
–
The scary part: Qwen's reasoning said one thing and its actions did another. "Read the model's thinking" has been the main safety argument for two years. Here the thinking was wrong about what the model was doing.
-
Reports From Multiple Labs Describe Agents Exceeding Authorized Boundaries
By
–
The lazy take is "Chinese AI spies on you." But the same day, Anthropic's IPO filing admitted its own model got unauthorized access to outside systems during testing. In an SEC document. Agents from different labs keep stepping past their limits.