GPT-5.5 & GPT-5.5 Pro Master Prompts (June 2026)

/* GPT-5.5 MASTER PROMPT
   VERSION: 1.0.0
   CAPABILITIES: Native Multimodality, Structured Output, Reasoning, Function Calling, Agents SDK */

**Model Configuration:**
┌─────────────────────────────────────────────┐
│ model: gpt-5.5 | gpt-5.5-pro | gpt-5.5-instant│
│ reasoning_effort: low | medium | high        │
│ response_format: text | json_schema          │
│ modalities: [text, vision, audio, speech]    │
│ tools: [custom_functions, code_interpreter]  │
└─────────────────────────────────────────────┘

**Prompt Architecture:**
┌─ SYSTEM ─────────────────────────────────────┐
│ Role: [EXPERTISE + CONSTRAINTS]              │
│ Output: [FORMAT_SPEC]                        │
│ Rules: [SAFETY + QUALITY_BOUNDARIES]          │
└──────────────────────────────────────────────┘
┌─ CONTEXT ────────────────────────────────────┐
│ Background: [DOMAIN_KNOWLEDGE]               │
│ Input Data: [TEXT | IMAGE | AUDIO | SPEECH]  │
│ Constraints: [HARD_REQUIREMENTS]             │
└──────────────────────────────────────────────┘
┌─ INSTRUCTION ────────────────────────────────┐
│ 1. [STEP_ONE] — purpose + expected output    │
│ 2. [STEP_TWO] — dependencies on step 1       │
│ 3. [STEP_N] — final integration              │
└──────────────────────────────────────────────┘

**Reasoning Mode Selection:**
- LOW: Simple classification, formatting, translation
- MEDIUM: Code review, content writing, data analysis
- HIGH: Architecture design, multi-step research, debugging complex systems

**Multimodal Input Pattern:**
```
[IMAGE: screenshot of dashboard]
Describe the metrics shown and identify anomalies.

[AUDIO: customer_call.mp3]
Extract action items and sentiment shifts.

[TEXT: product spec]
Given the above, generate a prioritized roadmap.
```

**Function Calling Pattern:**
Define tools with clear JSON schemas. GPT-5.5 will chain calls intelligently.
For agent workflows, use the Agents SDK pattern: plan → execute → verify → integrate.

**Output Quality Gates:**
- Structured outputs: validate against schema before returning
- Code: include error handling and edge cases
- Research: cite sources, flag uncertainty
- Creative: provide 3+ variations with trade-off notes