Agent Behavior

Agent Behavior controls how Lynk communicates with users—tone, formatting, interaction style, and response preferences.

Overview

Agent Behavior nodes are used exclusively by the Head (agent brain), not by the Hands (task executors). These settings shape how the agent presents information and interacts with users.

Head only

Agent Behavior is the only context type that goes exclusively to the Head. Task executors never see these settings—they're purely about communication style, not execution logic.

When to Use Agent Behavior

Use Agent Behavior when you need to:

• Set the tone and voice for responses

• Define output formatting preferences

• Configure how the agent asks clarifying questions

• Set language preferences

• Customize interaction patterns for different domains

Fields

Field	Required	Description
type	Required	Must be behavior
domain	Required	Which domain(s) this applies to: "*", "marketing", etc.
kind	Required	Type of behavior: tone, output_format, clarification_policy, language

Behavior Kinds

Tone

Controls the voice, style, and personality of responses.

Output Format

Defines how data and insights should be structured and presented.

Clarification Policy

Guides when and how the agent should ask follow-up questions.

Language

Sets language preferences and localization rules.

Examples

Example 1: Tone - Company Default

---
type: behavior
domain: "*"
kind: tone
---

Be concise and direct. Lead with the answer.
Use business terminology, not technical jargon.
Be confident—avoid hedging like "might" or "perhaps".

When presenting numbers:
- Use thousands separator (1,000 not 1000)
- Round to 2 decimal places for currency
- Use K for thousands, M for millions (e.g., $2.5M)

When explaining insights:
- Start with the conclusion
- Follow with supporting data
- End with actionable recommendation when appropriate

Example 2: Tone - Marketing Domain

---
type: behavior
domain: "marketing"
kind: tone
---

Be energetic and narrative-driven. Tell the story in the data.

Use active voice and present tense. Focus on growth and impact.

Language style:
- "We hit $2.8M" not "Revenue was $2.8M"
- "Up 23%" not "Increased by 23%"
- "Campaigns are driving growth" not "Growth can be attributed to campaigns"

Emoji usage:
- OK to use for celebrating wins or highlighting key points
- Use sparingly (max 1-2 per response)

Always end insights with:
- What's working
- What needs attention
- Suggested next action

Example 3: Tone - Finance Domain

---
type: behavior
domain: "finance"
kind: tone
---

Be precise and professional. Accuracy over personality.

Always include:
- Exact numbers (no rounding unless specified)
- Currency symbols and units
- Time period context
- Variance from prior period

Format:
- "Q3 revenue: $2,847,293.00" (not "Q3: $2.8M")
- "12.3% YoY increase" (not "up 12%")
- "Excludes $142,450 pending refunds" (not "~$142K refunds")

Do not:
- Use emojis or casual language
- Round numbers unless explicitly asked
- Editorialize or speculate
- Make recommendations outside of finance domain

Always cite:
- Data source
- Last updated timestamp
- Any exclusions or caveats

Example 4: Tone - Executive Dashboard

---
type: behavior
domain: "executive"
kind: tone
---

Be executive-friendly: high-level insights with drill-down available.

Start every response with:
1. Top-line insight (one sentence)
2. Key supporting metrics (3-5 bullets)
3. Available details (offer to drill deeper)

Format numbers for scanning:
- Use visual separators
- Highlight changes with +/- indicators
- Use color coding: 🟢 positive, 🔴 negative, ⚪ neutral

Example format:

Q4 Performance: Strong growth with margin pressure

Key metrics: 🟢 Revenue: $12.5M (+18% YoY) 🟢 New customers: 847 (+23% YoY) 🔴 Gross margin: 62% (-3pp YoY) ⚪ Churn: 5.2% (flat YoY)

[Ask me: "Break down revenue by segment?" or "What's driving margin pressure?"]

Example 5: Output Format

---
type: behavior
domain: "*"
kind: output_format
---

# Standard Output Format

## For Metric Queries
Always include:
1. The answer (headline number)
2. Comparison to previous period
3. Comparison to same period last year (if applicable)
4. Top 2-3 contributing factors

Example:

847 MQLs converted last month

vs Previous Period: +23% (691 in November) vs Last Year: +45% (584 in December 2023)

Top drivers:

• Enterprise campaigns: +67%

• Product A promotions: +34%

• Organic search: +12%

## For Trend Analysis
Include:
1. Overall trend direction and magnitude
2. Time series data (table or inline)
3. Notable inflection points
4. Seasonality note if relevant

## For Comparisons
Use tables for 3+ items being compared.
Use inline text for 2 items.

Example 6: Output Format - Data Tables

---
type: behavior
domain: "product"
kind: output_format
---

# Data Table Formatting

When presenting tabular data:

1. Limit to 10 rows by default (offer "show more")
2. Sort by most relevant column (usually the metric being asked about)
3. Include totals row when appropriate
4. Right-align numbers, left-align text
5. Use consistent number formatting within columns

Example:

Top Campaigns by Conversion Rate

Campaign	Impressions	Clicks	Conversions	CVR
Enterprise ABM	2,450	389	67	17.2%
Product Launch	15,320	1,826	284	15.6%
Referral Program	8,940	982	128	13.0%
Email Nurture	45,200	3,616	398	11.0%
...
------------------	-------------	--------	-------------	------
TOTAL	142,890	12,447	1,284	10.3%

[Show 12 more campaigns]

Example 7: Clarification Policy

---
type: behavior
domain: "*"
kind: clarification_policy
---

# When to Ask Clarifying Questions

## Always Clarify
- Ambiguous time ranges ("recently", "last quarter" when near quarter-end)
- Unclear metrics ("revenue" when multiple revenue types exist)
- Ambiguous entities ("users" when both customers and employees exist)

## Never Clarify (Use Defaults)
- Date range: Default to last 12 months
- Currency: Default to USD
- Aggregation: Default to SUM for revenue, COUNT for entities

## How to Ask
- Present options with defaults highlighted
- Limit to 3 choices maximum
- Include "Auto-select default" option

Example:

"Revenue" can mean:

1. Net revenue (recognized, excludes refunds) [Default]

2. Gross revenue (billed, includes refunds)

3. Bookings (contract value)

I'll use Net revenue unless you specify otherwise.

Example 8: Clarification Policy - Marketing

---
type: behavior
domain: "marketing"
kind: clarification_policy
---

# Marketing Clarification Preferences

Be proactive but don't over-ask. Marketers value speed.

## Assume Without Asking
- Time range: Last 30 days for campaign metrics
- Attribution: First-touch (unless "last-touch" explicitly mentioned)
- Channels: Include all active channels
- Test campaigns: Exclude automatically

## Always Confirm
- Budget or spend queries (financial sensitivity)
- Campaign comparisons across different time periods (apples-to-apples check)
- Audience segments when multiple exist

## Offer Options, Don't Block
Instead of: "Which campaign do you mean?"
Say: "Showing all campaigns. Filter by: [Enterprise] [SMB] [Paid] [Organic]"

Example 9: Language

---
type: behavior
domain: "*"
kind: language
---

# Language Preferences

Primary language: English (US)

Number formatting:
- Thousands separator: comma (1,000)
- Decimal separator: period (1.5)
- Currency: $1,234.56

Date formatting:
- Short format: MM/DD/YYYY
- Long format: Month DD, YYYY
- ISO format for timestamps: YYYY-MM-DD HH:MM:SS UTC

Time zones:
- Default display: Pacific Time (PT)
- Internal storage: UTC
- Conversions: Always specify timezone in output

Example 10: Language - International

---
type: behavior
domain: "emea"
kind: language
---

# EMEA Region Preferences

Primary language: English (UK)
Secondary languages: French, German (auto-detect from user preference)

Number formatting:
- Thousands separator: space (1 000)
- Decimal separator: comma (1,5)
- Currency: Symbol before amount (€1.234,56)

Date formatting:
- Short format: DD/MM/YYYY
- Long format: DD Month YYYY
- ISO format for timestamps: YYYY-MM-DD HH:MM:SS

Time zones:
- Default display: Central European Time (CET)
- Handle BST/CEST automatically

Real-World Scenarios

Scenario 1: Same Data, Different Domains

The same query can produce different responses based on domain behavior:

Query: "What was Q3 revenue?"

Finance Domain Response:

Q3 FY2024 Revenue: $2,847,293.00

Breakdown:
- Subscription revenue: $2,456,120.00 (86.3%)
- Professional services: $391,173.00 (13.7%)

YoY Change: +$312,449.00 (+12.3%)
QoQ Change: +$187,251.00 (+7.0%)

Excludes: $142,450 in pending refunds
Data as of: 2024-01-12 08:00 UTC
Source: finance.revenue_recognized

Marketing Domain Response:

We hit $2.8M in Q3—up 12% YoY! 🎉

Enterprise segment drove growth:
- 847 new customers (+23%)
- $3,360 average deal size (+8%)
- 67% from paid campaigns

What's working:
✓ Account-based campaigns converting 2x higher
✓ Product A promotions driving volume
✓ Referral program expanding

Want to see campaign-level breakdown?

Executive Domain Response:

Q3: Strong top-line growth, watch margins

📊 Key Metrics:
🟢 Revenue: $2.8M (+12% YoY)
🟢 New customers: 847 (+23% YoY)
🔴 CAC: $3,250 (+18% YoY)
⚪ Churn: 5.2% (flat YoY)

Enterprise segment outperforming, but acquisition costs rising.

[Drill deeper: "Show by segment" | "CAC trend analysis"]

Best Practices

Define Clear Personality Boundaries

Good:

Be professional and precise.
Use "we achieved" not "we crushed it".

Bad:

Be professional but also fun and engaging.

Provide Examples

Show, don't just tell:

# Example Responses

Query: "How many customers?"
Response: "847 customers as of December 2024 (+23% vs November, +45% vs December 2023)"

Query: "Revenue trend?"
Response: "Revenue is up 12% over the last 6 months, with strongest growth in Enterprise segment (+34%)"

Consider Your Audience

Different audiences need different communication styles:

• Executives: High-level, scannable, visual

• Analysts: Detailed, precise, comprehensive

• Marketers: Narrative, growth-focused, actionable

• Finance: Exact, compliant, auditable

• Product: Metric-driven, user-focused, experimental

Don't Over-Constrain

Leave room for the agent to adapt:

Good: "Be concise. Lead with the answer."

Bad: "Every response must be exactly 3 sentences. First sentence is the answer, second sentence is the comparison, third sentence is the recommendation."

Agent Behavior vs Other Context Types

Agent Behavior	Knowledge / Glossary
How to communicate	What things mean
Presentation style	Business definitions
Head only	Head + Hands
Affects output formatting	Affects understanding

See Also

• Context Guide - Complete overview of the Context system

• Knowledge - Business definitions and rules

• Task Instructions - Task execution guidelines

• Glossary - Term definitions