Preemptive output control.
Restraint embedded at the point of first response — before the system commits to action.
Shard Protocol is an AI prompt engineering framework by Marcos Rezende. It fragments your instruction into six semantic shards — Goal, Context, Format, Constraints, Risk, and Reasoning — then composes a validated, structured prompt locally in your browser.
How it works
- Raw Input Provide an unstable or ambiguous instruction.
- Parse & Segment The protocol fragments it into six semantic shards.
- Apply Controls Risk tolerance and format settings are applied per shard.
- Structure Output The local composer creates a safe, validated prompt.
- Refine or Use Copy the final structured prompt or iterate further.
Each step ensures preemptive control before AI execution — analyze intent, identify risks, apply constraints, and ensure safety before the model ever responds.
The six shards
- Goal — The core objective the AI must achieve.
- Context — Background information that frames the task.
- Format — Required output structure and style.
- Constraints — Hard limits the AI must not cross.
- Risk — Potential failure modes identified upfront.
- Reasoning — The chain-of-thought logic the AI should follow.
Input Fragmentation
Enter any AI instruction. Shard Protocol analyzes it for safety and structure, breaks it down into components, checks for potential risks, and produces a preemptively controlled prompt ready for use with any large language model.