Agent Test Skill ⭐

The most powerful way to use FluxLoop. Just talk naturally—the skill handles everything.

Why Use the Skill?

Approach	Experience
Skill (Recommended)	"test my agent" → Everything happens automatically
Slash Commands	Manual: /fluxloop:pull → /fluxloop:test → check results

The skill is context-aware. It:

• Knows your current setup state
• Guides you through missing steps
• Asks before executing anything
• Handles errors gracefully

---

How to Use

Just ask Claude naturally:

"test my agent"
"generate test data"
"run a simulation"
"set up fluxloop"
"evaluate my chatbot"

That's it. The skill activates automatically.

---

Activation Keywords

The skill activates when you mention:

Category	Keywords
Testing	test, testing, test my agent, run test
Simulation	simulation, run simulation, simulate
Data Generation	generate test data, synthesize, create inputs
Evaluation	evaluate, evaluation, assess, check quality
Setup	set up fluxloop, configure, initialize

---

Skill Workflow

Core Principle: Context-First

The skill always follows this pattern:

1. Check context (fluxloop context show)
      ↓
2. Summarize current state to you
      ↓
3. Present options (NO auto-execution)
      ↓
4. Execute after your confirmation
      ↓
5. Show results and next steps

You're always in control. Nothing runs without your approval.

---

Example Conversations

First-Time Setup

You: "I want to test my agent with fluxloop"

Claude: I'll help you set up FluxLoop for agent testing.

Let me check your current state...

📊 Current State:
- FluxLoop CLI: ✗ Not installed
- Authentication: -
- Project: -

To get started, I need to:
1. Install FluxLoop CLI
2. Log you in
3. Create or select a project

Should I proceed with the installation?

You: "yes"

Claude: Installing FluxLoop CLI...
[Executes: pip install fluxloop-cli]

✓ FluxLoop CLI installed!

Now let's log you in. Run this command and follow the browser prompt:

[Executes: fluxloop auth login]

...

Generate Test Data

You: "generate test data for my customer support bot"

Claude: I'll generate test data for your agent.

📊 Current State:
- Project: customer-support (proj_abc)
- Scenario: support-bot (scn_123)
- Existing bundles: 2 found
  - v2 (stress-test, 50 inputs, 1 day ago)
  - v1 (basic, 10 inputs, 1 week ago)

Would you like to:
1. Use existing bundle v2 (50 inputs)
2. Use existing bundle v1 (10 inputs)  
3. Generate new test data

You: "3"

Claude: I'll generate new test data. First, let me create personas...

[Executes: fluxloop personas suggest --scenario-id scn_123]

✓ 5 personas generated:
- frustrated_customer
- new_user
- power_user
- confused_elderly
- impatient_professional

Now synthesizing test inputs...

[Executes: fluxloop inputs synthesize --scenario-id scn_123]

✓ 30 test inputs generated!

Publishing as bundle v3...

[Executes: fluxloop bundles publish ...]

✓ Bundle v3 published! Ready to test.

Run tests now?

Run Tests

You: "run the tests"

Claude: Running tests with bundle v3...

[Executes: fluxloop sync pull --bundle-version-id bv_xyz]
[Executes: fluxloop test --scenario support-bot]

📊 Test Results:
✓ Passed: 27/30 (90%)
⚠️ Warnings: 3

Failed cases:
1. [frustrated_customer] "This is ridiculous!" 
   → Agent used "sorry" (forbidden word)
   
2. [power_user] "What's the API rate limit?"
   → Response too long (2500 chars > 2000 limit)
   
3. [impatient_professional] "Just fix it now"
   → Empty response

🔗 Detailed analysis: https://app.fluxloop.ai/runs/...

Would you like me to analyze these failures and suggest fixes?

---

State-Based Guidance

The skill adapts to your current state:

Your State	Skill Response
No CLI installed	Installs CLI automatically
Not logged in	Helps with authentication
No project	Lists projects or creates new
No scenario	Creates local scenario folder
No test data	Generates personas and inputs
Ready to test	Runs tests immediately

---

Decision Tree

When you ask to test, the skill follows this logic:

Check bundles list
  │
  ├─ Multiple bundles → "Which bundle?" (shows list)
  │
  ├─ One bundle → "Use existing or create new?"
  │
  └─ No bundles → Check inputs
                    │
                    ├─ Has inputs → "Publish as bundle?"
                    │
                    └─ No inputs → "Generate new data?"

---

Key Commands Used by Skill

The skill orchestrates these CLI commands for you:

Phase	Commands
Setup	fluxloop auth login, fluxloop projects select
Scenario	fluxloop init scenario, fluxloop scenarios create
Data Gen	fluxloop personas suggest, fluxloop inputs synthesize
Publish	fluxloop bundles publish
Test	fluxloop sync pull, fluxloop test
Results	fluxloop test results

You don't need to remember any of these. Just talk naturally.

---

Best Practices

1. Be Specific About Your Agent

Good: "test my customer support chatbot"
Good: "generate edge cases for my order cancellation flow"

Less helpful: "test it"

2. Let the Skill Guide You

Don't try to run commands manually. Say what you want to achieve:

Good: "I want to test with more difficult scenarios"
Less helpful: "/fluxloop:synthesis" (manual command)

3. Ask for Analysis

After tests run, ask:

"Why did test #3 fail?"
"How can I fix these warnings?"
"What patterns do you see in the failures?"

---

Troubleshooting

Skill Not Activating?

Make sure your message includes activation keywords:

✗ "check this"
✓ "test this agent"

✗ "make some data"  
✓ "generate test data"

Wrong Context?

If the skill is checking the wrong project:

"switch to project customer-support"
"use scenario order-bot"

Need Manual Control?

For edge cases, you can use slash commands:

/fluxloop:status   → Check current state
/fluxloop:pull     → Pull specific bundle
/fluxloop:test     → Run with specific options

But 99% of the time, just talk naturally!

---

Related

• Installation - Plugin setup
• Commands Reference - Manual commands
• Workflow Integration - Development workflow
• FluxLoop Web - Cloud dashboard