🤖 Agent Design Canvas — AgentSea Edition

Pick a real workflow from your team, design the agent, write the system prompt, and deploy it on AgentSea.

⏱ 60 minutes

Exercise Overview

This is the capstone exercise of the workshop. You've learned what AI can do, how to communicate with it effectively, and what AgentSea's skills can handle. Now you'll put it all together — designing an AI agent for a real workflow from your own team.

You'll work in teams of 3-4 to select a workflow, design the agent using the canvas template, write a deployable system prompt, and test it on AgentSea. Your final deliverable is a system prompt that can be deployed as an agent in the AgentSea marketplace.

🎯 Deliverable

Each team produces a System Prompt for AgentSea — designed using the Agent Design Canvas. Submit it for AI scoring on the leaderboard.

Step	What you do	Duration
Step 1	Pick your workflow	5 min
Step 2	Fill the Agent Design Canvas	20 min
Step 3	Write the system prompt	15 min
Step 4	Test on AgentSea platform	10 min
Step 5	Submit for AI scoring	10 min

Step 1: Pick Your Workflow

Click a workflow below to select it. Choose one that your team does at least weekly and takes at least 30 minutes each time. Or pick your own!

🏥 Vendor Proposal Analysis

Read proposal → extract terms → compare requirements → recommend

Skills: Document Analysis

📋 Meeting Minutes & Action Items

Read notes → extract decisions → assign owners → draft follow-up

Skills: Doc Analysis + Doc Gen + Outlook

💰 Budget Variance Report

Read financials → identify variances → explain → generate report

Skills: Doc Analysis + Doc Gen

📧 Email Triage & Response

Read inbox → classify urgency → draft responses → flag priorities

Skills: Outlook + Doc Gen

📄 Policy Compliance Check

Read policy + regulation → compare → flag gaps → action items

Skills: Doc Analysis + Doc Gen

🔄 Leave & Claims Processing

Read form → validate policy → check entitlement → recommend

Skills: Doc Analysis + Doc Gen

📊 Monthly KPI Narrative

Read KPIs → analyze trends → flag misses → generate commentary

Skills: Doc Analysis + Doc Gen

🏗️ Project Status Consolidation

Read updates → extract status → consolidate → flag risks

Skills: Doc Analysis + Doc Gen

⚠️ Best choice: Pick a workflow that your team does at least weekly and that takes at least 30 minutes each time. The bigger the time savings, the stronger your business case. You can also design your own workflow — just describe it in Step 2.

Step 2: Fill the Agent Design Canvas

Use this canvas to think through every aspect of your agent. The canvas is pre-filled based on your workflow selection — review and customize it for your team's specifics.

AGENT DESIGN CANVAS — Fill this out with your team

# Agent Design Canvas — AgentSea Edition ## Agent Identity - **Name:** [Give your agent a descriptive name, e.g., "Procurement Analyst Agent"] - **Role:** [Use persona formula: Title + Experience + Specialty + Behavior] Example: "Senior Procurement Analyst with 10 years of healthcare vendor assessment experience, methodical and thorough" - **Purpose:** [One sentence: what does this agent help you do?] ## Workflow - **Trigger:** [When do you use this agent? e.g., "When I receive a new vendor proposal"] - **Steps:** 1. [What do you give the agent? (document, email, data)] 2. [What should the agent do with it? (analyze, summarize, generate)] 3. [What output do you expect? (report, email draft, checklist)] ## Skills Required - [ ] Document Analysis (read PDFs, Excel, Word) - [ ] Document Generation (create reports, letters, summaries) - [ ] Outlook Integration (read emails, calendar) ## Data & Context - **Input documents:** [What files will you upload each session?] - **Reference knowledge:** [What policies/guidelines should the agent know?] - **Output format:** [Table? Bullet points? Email draft? PDF report?] ## Guardrails (System Prompt Rules) - **Must always:** [e.g., "cite the source document", "use formal tone"] - **Must NEVER:** [e.g., "make up data", "provide medical advice", "share PII"] - **Escalate when:** [When should the agent say "please check with your supervisor"?] ## Business Impact - **Current process:** [How is this done today? How long does it take?] - **With agent:** [Expected time savings per use] - **Frequency:** [How often do you do this task?] - **Weekly time saved:** [Calculate: time saved × frequency]

💡 Tips for a strong canvas:

Be specific about the trigger — "when I receive a new vendor proposal PDF" is better than "when needed"
Name the output format — "a table with 5 columns" is better than "a summary"
Include escalation rules — when should a human take over?
Quantify the business impact — "saves 2 hours per proposal × 4 proposals/week = 8 hours/week"
Use negative constraints — what the agent must NEVER do is as important as what it should do

Step 3: Write the System Prompt

Convert your canvas into a deployable system prompt. This is what you'll paste into AgentSea to create your agent. Use the template below as a starting point:

SYSTEM PROMPT TEMPLATE — Customize for your agent

You are [AGENT NAME] — [ROLE with persona formula]. Your purpose is to [ONE SENTENCE from canvas]. ## What You Do When the user uploads [INPUT TYPE from canvas], you will: 1. [STEP 1 from canvas — what to extract/analyze] 2. [STEP 2 from canvas — what to produce/generate] 3. [STEP 3 from canvas — what format to output] ## Output Format [Describe the exact format from your canvas: table columns, bullet structure, report sections] ## Rules You MUST Follow - Only use information from the uploaded document(s) — never fabricate data - If information is missing, state "NOT PROVIDED" — never guess or assume - All monetary values in SGD - Use professional, clear language appropriate for [AUDIENCE] - [ADD MORE from your "Must always" guardrails] ## You Must NEVER - [From canvas: Must NEVER item 1] - [From canvas: Must NEVER item 2] - [From canvas: Must NEVER item 3] - Disclose any patient identifiable information - Provide medical, legal, or financial advice ## When to Escalate Say "⚠️ This requires human review — please consult your supervisor" when: - [From canvas: Escalation trigger 1] - [From canvas: Escalation trigger 2] - You are unsure about the correct interpretation - The request falls outside your defined scope

✅ Checkpoint: Your system prompt should be:

Specific to your workflow (not generic)
Include at least 3 "Must NEVER" rules
Include at least 2 escalation triggers
Define a clear output format
Ready to paste into AgentSea

Step 4: Test on AgentSea

Deploy your system prompt on the AgentSea platform and test it with a sample document.

Open AgentSea — go to the platform and create a new agent
Paste your system prompt — in the agent configuration, paste the system prompt from Step 3
Select skills — enable the skills you identified in your canvas (Document Analysis, Document Generation, Outlook)
Upload a test document — use a sample document that represents your typical workflow input
Test the interaction — ask the agent to process the document and check the output

🧪 Testing checklist:

Does the output match the format you specified?
Does the agent cite the source document?
Try asking something outside scope — does it escalate correctly?
Try giving incomplete data — does it say "NOT PROVIDED"?
Is the tone appropriate for your audience?

Step 5: Submit & Score

Submit your system prompt for automated AI scoring. The AI judge evaluates your agent design on 6 criteria.

📤 Submit Your System Prompt

Team / Name Your System Prompt (paste the full system prompt)

💡 How scoring works

Your system prompt is evaluated by AI on these criteria:

Criteria	What it measures	Score
Clarity & Specificity	Is the role, purpose, and workflow clearly defined?	/5
Guardrails & Safety	Are there strong "must never" rules and escalation triggers?	/5
Output Structure	Is the expected output format well-defined?	/5
Persona & Tone	Does the persona formula create a believable expert?	/5
Business Impact	Is the time savings quantified and realistic?	/5
Healthcare Relevance	Is it grounded in Synapxe's healthcare admin context?	/5

25-30: 🥇 Production-Ready | 19-24: 🥈 Strong Design | 13-18: 🥉 Good Start | Below 13: 🔄 Needs Rework

🏆 Leaderboard

All team submissions ranked by score. Click "Refresh" to see new entries.

No submissions yet. Be the first to submit!

← Workshop Home Prompt Exercises →