> For the complete documentation index, see [llms.txt](https://alludium.gitbook.io/alludium-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://alludium.gitbook.io/alludium-docs/administration/3.-agent-builder/3.2-prompt-and-instructions.md).

# Prompt and Instructions

To access prompt configuration, open your agent in Agent Builder, click **Configuration**, then select the **Prompt** sub-tab.

The system prompt is the agent's operating logic. It defines role, behavioral constraints, output standards, and governance rules.

When writing agent instructions, treat configuration as product design — not prompt writing. A well-configured agent specifies:

**Purpose and scope** — What problem does this agent solve? When should it be invoked? When should it explicitly not be used?

**Inputs** — What context does the agent need? Define this explicitly: workspace files, uploaded documents, structured notes from the user, or project context. Agents produce better outputs when context is complete and deliberate — never assume the agent will infer what it needs.

**Output definition** — What does the primary output look like? Define format (document, table, list, summary), level of polish (working draft vs. near-final), and intended next step (review, approval, handoff to another agent).

**Guardrails** — What should the agent not do? Examples: no recommendations without supporting rationale, no speculative claims without evidence, no deletion or archiving of source documents.

***

### Prompt Structure

Effective prompts follow this pattern:

**1. Role definition**\
Who is this agent? What expertise does it have?

**2. Primary task**\
What does the agent do?

**3. Input requirements**\
What information does the agent need to complete its task?

**4. Output specification**\
What format, structure, and level of detail should the output have?

**5. Constraints and guardrails**\
What should the agent never do? What are the boundaries?

**6. Tone and voice**\
How formal, technical, or conversational should outputs be?

***

### Example Prompt

```
You are a research agent focused on early-stage B2B SaaS companies.

Your task is to build structured company profiles based on public information.

Required inputs:
- Company name
- Company website URL
- Target information depth (brief, standard, comprehensive)

Output structure:
- Company Overview (2-3 paragraphs)
- Business Model (revenue model, target customers, key value proposition)
- Funding History (rounds, investors, amounts, dates)
- Key Personnel (founders, executives with LinkedIn profiles)
- Recent News (past 90 days, 3-5 items with dates and sources)

Constraints:
- Always cite sources for all claims
- If information is unavailable, state that explicitly rather than speculating
- Do not include subjective assessments or investment recommendations
- Do not contact the company directly or scrape non-public information

Tone: Neutral and analytical. Use business language appropriate for investment professionals.
```

***

### Testing Your Prompt

Before deploying, test with these scenarios:

**Standard case** — Typical inputs, expected workflow\
**Edge case** — Unusual inputs, boundary conditions\
**Missing information** — What happens when required data isn't available?\
**Ambiguous request** — Does the agent ask clarifying questions or make assumptions?

***

### Iterating Prompts

After deployment, refine based on:

**Output quality** — Are outputs meeting standards? Do they require heavy editing?\
**Scope creep** — Is the agent attempting tasks outside its defined purpose?\
**User confusion** — Are users unclear about what inputs to provide?\
**Error patterns** — Does the agent consistently mishandle certain scenarios?

Edit the agent, update the prompt, test changes, redeploy. Monitor the next 5-10 invocations for improvement.

***

### Common Mistakes

❌ **Vague purpose:** "Help with research"\
✅ **Specific purpose:** "Profile B2B SaaS companies for investment screening"

❌ **Assumed inputs:** "Research the company"\
✅ **Explicit inputs:** "Required: company name, website URL. Optional: specific focus areas"

❌ **Undefined outputs:** "Provide a summary"\
✅ **Structured outputs:** "5-section profile: Overview, Model, Funding, Team, News"

❌ **Missing guardrails:** (no constraints specified)\
✅ **Clear boundaries:** "Never make investment recommendations. Never speculate beyond public information."

***

### Next Steps

With prompts and instructions configured, continue to **AI Model Configuration** to select and tune the underlying model for optimal performance.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://alludium.gitbook.io/alludium-docs/administration/3.-agent-builder/3.2-prompt-and-instructions.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.