Source: base_instructions field for slugs gpt-5.5 and gpt-5.4 in models_cache.json.
Both prompts share the same skeleton. Personality, General, Editing constraints, Special user requests, Autonomy and persistence, Working with the user, Formatting, Final answer, Intermediary updates. The 5.5 prompt is ~46% longer (21.5 KB vs 14.7 KB) and the additions are not neutral, they materially change what the model is told to do before it touches code.