Find where your prompt breaks before your users do.
prompt.txt
Act as a QA engineer for prompts. The prompt I plan to use in production: [paste prompt]. Its job: [what it should produce]. Who will use it: [audience]. What a failure costs me: [stakes].
Stress-test it like a hostile reviewer:
1. Write 5 realistic inputs that could break it: one vague, one adversarial, one missing key info, one overloaded with detail, one completely off-topic.
2. For each input, predict the most likely failure mode and rate severity from 1-5 with one line of reasoning.
3. Rewrite my prompt to survive all five, keeping it under [word limit] words. Show the rewrite in full.
4. End with a 3-line pre-flight checklist I can paste above the prompt to catch bad inputs early.
#prompting#testing
💡 Replace the [brackets] with your details — the more specific, the sharper the answer.
Sort a crowded inbox without losing the messages that matter.
Act as a practical specialist helping me with inbox triage blueprint. My context: [mail volume, categories, urgent senders, tools available].
1. Define priority tiers and exact inclusion rules.
2. Write example labels and false-positive cases.
3. Design a daily review queue and escalation path.
4. Give a five-message test set with expected labels..
Use my details, not a generic example. If a fact needed for the plan is missing, ask for it or mark the assumption. Keep the result ready to use.
Turn a broad question into a source-checked investigation.
Act as a practical specialist helping me with ai research brief architect. My context: [question, decision it supports, deadline, sources allowed].
1. Break the question into falsifiable subquestions.
2. Set a source hierarchy and recency requirements.
3. Create an evidence table with claim, source, date, confidence.
4. Specify when to stop researching and how to flag gaps..
Use my details, not a generic example. If a fact needed for the plan is missing, ask for it or mark the assumption. Keep the result ready to use.
Make messy documents yield clean, auditable fields.
Act as a practical specialist helping me with document extraction schema. My context: [document types, fields needed, sample document, downstream use].
1. Design a JSON schema with field types and null handling.
2. Write extraction rules with exact quote or page evidence.
3. List ambiguous cases and a human-review threshold.
4. Produce three test cases including a missing field..
Use my details, not a generic example. If a fact needed for the plan is missing, ask for it or mark the assumption. Keep the result ready to use.