Shrutam CCA-F Crash › Domain 4 › d4-l03-system-prompt-design Hinglish English →

System Prompts — Role, Rules, Constraints

Domain 4 · 18% ~12 min Hinglish narration

Audio-only (commute / mobile data)

Same Saavi narration, smaller file. Opus 48k preferred — auto-selected by your browser.

Scenario anchor

Aaranya IT BFSI team ne ek Claude-powered loan-underwriting assistant deploy kiya — lekin production mein pehle hi hafte, model ne compliance-mandated JSON fields drop karne shuru kar diye, kyunki user prompts system instructions ko override kar rahe the. Aap us senior architect hain jinhone yeh escalation receive ki hai. Is lesson ke baad aap precisely jaanenge ki system prompt ko kaise design karein taaki role, rules, aur output constraints production-grade enforcement ke saath hold karein — chahe user input kuch bhi bheje.

Key Takeaways

System prompt is the authoritative config layer — role defines identity, rules define permitted operations, and output constraints define the schema contract; all three must be explicit and co-located in the system turn, not scattered across user turns.
Instruction hierarchy in Claude flows system → human → assistant; any constraint not declared at system level is subject to user-turn override, which is a production-grade security and compliance risk in regulated domains like BFSI or legal.
Output schema contracts belong in the system prompt as typed, enumerated specifications — treat them the same way aap apne OpenAPI spec mein request/response schemas enforce karte hain, not as suggestions.
Memory anchor: system prompt is your config-cascade root node — everything downstream inherits or is bounded by what is declared here; if it is not in the config, it is not enforced.

← Previous lesson Same lesson in English Next lesson → Back to overview