The Instructions section is where you define the Enterprise Search agent's identity and behavior: "Define how your Enterprise Search Agent should behave and what users see when starting a conversation." It sits under Enterprise Search Agent behavior, and it's the same editor an AI agent uses — worded here for a grounded, cite-your-sources search assistant.

What is the Instructions section?

You fill three structured fields — a Role, a Goal, and the long-form Instructions body — plus the conversation starters users see when they first open the agent. This is distinct from a Prompts section that might hold named, reusable templates; Instructions is the agent's core, always-on guidance.

How do I set the agent's role and goal?

The screen opens with two labeled, single-line boxes:

Keep them concrete: together they frame who the agent is and what it's for, and they anchor everything else it does.

How do I write the agent's instructions?

Below Role and Goal is the Instructions field, previewed with the hint "Detailed rules for tone, format, and behavior." Click it, or the pencil icon, to open a full-screen editor headed "Provide background context and instructions related to your agent's expertise."

The Enterprise Search placeholder is a ready-made grounded-answer policy: answer only from indexed content the user can access, cite the source for every claim, prefer the most recent document when sources conflict, say so when nothing relevant is indexed, and ask a clarifying question when a term is ambiguous.

The editor has a Preview / Code toggle and a floating toolbar to Add variable (inserts a {{ }} placeholder) and Add code (inserts a {% %} block); variables show as pills in both views. Click Done to bring your text back to the section — there's no enforced length limit, but concise, well-ordered instructions produce more reliable behavior than long ones.

The editor's more menu

A row labeled Add Instructions sits above the text and carries a more (three-dots) menu, a copy button once there's content, and the Preview/Code toggle. The menu groups:

• Upload instruction file, Download instructions, View diff — import a .md or .txt file (it overwrites the current instructions, so download a copy first), save the current text as instructions.md, or compare unsaved edits against the last saved version.

• Copy instruction, Clear all, Discard & close — copy the text, clear the field, or close without keeping changes. Copy instruction and Clear all appear only when there's content. Upload instruction file, Clear all and Discard & close are hidden when the instructions are read-only.

How do my instructions become the prompt the model sees?

At runtime, your role, goal, and Instructions body are compiled into the agent's system prompt using a template matched to the model family, so the same instructions work across models. The runtime then appends a few things you don't write by hand:

• Language and style directives from response-generation settings.

• Content-filter and word-filter guidance from guardrail configuration.

• Optionally, the output of a custom "additional context" automation from pre-processing, appended after it runs with the user's query.

• A hidden context message carrying the current date and time plus any per-user context, so answers stay time-aware.

Your instructions recompile fresh on every reasoning step, so edits take effect on the next turn.