o1/o3 reasoning prompt template

Prompt Template
System Message (or “Role” in some interfaces)

You are an expert [DOMAIN/ROLE], with deep analytical skills. You will carefully analyze any problem or scenario provided to you. Follow these rules:

1. Respond accurately and objectively, using only the given information plus common knowledge.
2. If relevant, identify assumptions or ambiguities before concluding.
3. Adhere to the requested output format and style.
4. Do not include your hidden chain-of-thought in the final answer unless explicitly asked.

Explanation:

Declare the model’s role or domain expertise (e.g., “legal analyst,” “senior law clerk,” “math teacher,” etc.).
Keep the instructions simple and direct.
If you want more or fewer details in the final answer, you can adjust item #4.

User Message (the main query)

Step 1: Provide Context (if any)

BACKGROUND/CONTEXT:
- [LIST OR SUMMARIZE ONLY THE RELEVANT FACTS, DATA, OR INFORMATION THE MODEL NEEDS]
- [INCLUDE ANY SPECIFIC STATUTES, CLAUSES, FORMULAS, ETC. THAT THE MODEL MUST USE IF APPLICABLE]

(Keep this section concise; avoid unnecessary detail.)

Step 2: State the Task/Question Clearly

TASK/QUESTION:
- [DESCRIBE THE EXACT QUESTION OR PROBLEM THE MODEL SHOULD SOLVE OR ANALYZE]
- [BE AS SPECIFIC AS POSSIBLE ABOUT WHAT YOU WANT ANSWERED]

Step 3: Specify Output Requirements

OUTPUT FORMAT & STYLE:
- [DESCRIBE THE DESIRED RESPONSE FORMAT: BULLET POINTS, IRAC (ISSUE-RULE-ANALYSIS-CONCLUSION), A TABLE, JSON, ETC.]
- [INDICATE VERBOSITY: E.G. "KEEP IT BRIEF" OR "PROVIDE A DETAILED STEP-BY-STEP EXPLANATION"]
- [OPTIONAL: IF YOU’RE USING O3-MINI AND HAVE ACCESS TO THE REASONING EFFORT PARAMETER, SPECIFY THE DESIRED REASONING LEVEL (LOW, MEDIUM, HIGH) HERE OR IN YOUR API SETTINGS]

(Optional) Additional Constraints
- [ANY SPECIAL INSTRUCTIONS, E.G., "DO NOT MAKE UP CITATIONS" OR "ONLY USE THE GIVEN FACTS"]
- [IF AMBIGUOUS, INSTRUCT: "STATE ANY ASSUMPTIONS YOU NEED TO MAKE"]
- [IF SPEED MATTERS: "PLEASE PROVIDE A DIRECT ANSWER WITHOUT EXCESSIVE DETAIL"]

(Optional) Follow-Up / Verification

- [ASK THE MODEL TO DOUBLE-CHECK OR VERIFY ITS ANSWER, IF CRUCIAL: E.G. "ONCE YOU REACH A CONCLUSION, REVIEW IT FOR CONSISTENCY AND POSSIBLE ERRORS."]