Skip to main content

Best Practices

Tips for using FluxLoop effectively with the Agent Test Skill.

Let the Skill Guide You​

✅ Do This​

"test my agent with difficult customers"
"generate more test data"
"what went wrong in the last test?"

❌ Avoid This​

/fluxloop:synthesis --count 50  (manual, misses context)
fluxloop test --scenario x      (raw CLI, no guidance)

The skill is context-aware. It knows your state and adapts.

Be Specific About Goals​

✅ Good Requests​

"test my order cancellation flow"
"generate edge cases for payment failures"
"run stress tests with impatient users"

❌ Vague Requests​

"test it"
"make some data"
"run tests"

The more context you give, the better the results.

Use Incremental Testing​

Development Loop​

1. Make small change
2. "quick smoke test"           → 5-10 inputs, fast feedback
3. Iterate until smoke passes
4. "run full test suite"        → Complete validation
5. Commit

Don't Skip Smoke Tests​

✗ Change code → Full test (slow, frustrating)
✓ Change code → Smoke test → Full test (fast iteration)

Ask for Analysis​

After tests run, ask:

"why did test #3 fail?"
"what patterns do you see in the failures?"
"how can I improve my agent's responses?"
"compare this run to yesterday's"

The skill can provide insights, not just results.

Maintain Test Data Quality​

Regenerate Periodically​

"generate fresh test data with current personas"

Old test data may not cover new edge cases.

Use Targeted Personas​

"add a persona for users who speak broken English"
"create edge cases for timezone issues"

Version Your Bundles​

"save this as bundle v4 with tag 'pre-launch'"

Team Collaboration​

Share Bundles​

"use the team's latest bundle"
"what bundles are available?"

Document Scenarios​

In FluxLoop Web, add descriptions to scenarios so team members understand what each tests.

Review Together​

"show me the test results from today"
"compare my results with the team average"

Performance Tips​

For Large Test Sets​

"run tests in batches of 10"
"test only the payment scenarios"

For Slow Agents​

"increase timeout to 60 seconds"
"run async tests"

For Quick Feedback​

"just run 3 random tests"
"smoke test only"

Troubleshooting​

Tests Failing Unexpectedly?​

"show me the test configuration"
"what changed since the last successful run?"

Skill Not Understanding?​

Be more explicit:

✗ "check it"
✓ "run fluxloop tests on my agent"

Need Manual Control?​

Fall back to commands:

/fluxloop:status
/fluxloop:test --scenario specific-scenario

Summary​

DoDon't
Use natural languageMemorize commands
Be specific about goalsMake vague requests
Run smoke tests firstJump to full tests
Ask for analysisJust look at pass/fail
Regenerate test dataUse stale data forever

Next Steps​