Tutorial 2: First Verification

March 31, 2026 · View on GitHub

This tutorial guides you through running your first verification and saving the results. You'll learn what happens during verification and how to practice with test fixtures.

Time: ~10 minutes
Prerequisites: Agent Verifier installed (Tutorial 1)

What You'll Learn

Run a full verification on your project
Understand what happens during verification
Save reports to files
Practice with test fixtures

Part 1: Run Your First Verification

Step 1: Open your project

Navigate to any AI agent project in your coding agent. If you don't have one handy, you can use the test fixtures (see Part 4).

Step 2: Trigger verification

In your coding agent, say:

"verify agent"

Alternative trigger phrases that also work:

"verify my agent"
"audit agent"
"full verification"

Step 3: Watch the verification process

The agent will work through several stages:

1. Context Discovery
   ├── Detecting project language...
   ├── Checking for agent framework...
   └── Looking for Kahuna integration...

2. Running Security Checks...
3. Running Pattern Checks...
4. Running Quality Checks...
5. Running Language-Specific Checks...
6. Consolidating Report...

Step 4: Review the summary

At the end, you'll see a summary like:

✅ 12 checks passed | ⚠️ 3 warnings | ❌ 1 issue

Part 2: Understanding the Verification Process

Agent Verifier runs four skill checks in sequence:

1. Security Checks (`verify-security`)

Scans for:

Hardcoded API keys and secrets
Unpinned dependencies
Missing input validation
Error messages that expose internals

2. Pattern Checks (`verify-patterns`)

Validates agent-specific patterns:

Loop safety: Do loops have termination conditions?
Retry limits: Are retry mechanisms bounded?
Tool consistency: Do prompts reference tools that exist?
Context size: Is the system prompt within token limits?

3. Quality Checks (`verify-quality`)

Reviews code quality:

Naming conventions
Code organization
Documentation coverage
Magic numbers/strings

4. Language Checks (`verify-language`)

Applies language-specific rules:

Python: Type hints, docstrings, requirements pinning
TypeScript: Strict mode, no any types, async error handling
Go: Error handling, context propagation

Part 3: Saving Reports

After verification completes, the agent will ask:

Would you like to save this verification report to a file?

Saving to default location

Say "yes" to save. The report is saved to:

reports/verification/YYYY-MM-DD_HH-MM-SS.md

Example: reports/verification/2026-03-27_14-30-45.md

Custom save location

You can specify a custom path:

"Save the report to docs/audit-report.md"

Report contents

The saved report includes:

Project metadata (name, date, language, framework)
Summary counts (pass/warn/issue)
Category breakdown table
Detailed findings with file locations
Prioritized recommendations

Part 4: Practice with Test Fixtures

Agent Verifier includes test fixtures to help you understand what it detects.

Navigate to fixtures

cd tests/fixtures

Run verification on all fixtures

In your coding agent:

"verify agent"

Expected results

The fixtures demonstrate various patterns:

Fixture	What it tests	Expected result
`infinite_loop/`	Loop termination detection	⚠️ Warnings for unbounded loops
`retry_limits/`	Retry limit enforcement	❌ Issues for missing limits
`tool_registry/`	Tool consistency	❌ Issue for hallucinated tool
`prompt_size/`	Context size awareness	✅ Pass (within limits)
`langgraph_cycles/`	Graph cycle analysis	❌ Issue for infinite cycle

Test individual categories

For focused testing:

cd tests/fixtures/retry_limits

Then say:

"verify agent patterns"

This runs only the pattern checks on the retry limit fixtures.

Part 5: Quick Reference

Trigger phrases

Phrase	What it runs
`"verify agent"`	Full verification (all 4 checks)
`"verify my agent"`	Full verification
`"audit agent"`	Full verification
`"verify agent security"`	Security checks only
`"verify agent patterns"`	Pattern checks only
`"verify agent quality"`	Quality checks only
`"verify agent language"`	Language checks only

What gets detected automatically

Language: Python, TypeScript/JavaScript, Go
Framework: LangGraph, CrewAI, AutoGen, LangChain, or custom
Kahuna: Organizational rules if .kahuna/ directory exists

Common Questions

How long does verification take?

Depends on project size:

Small projects (< 20 files): 30 seconds - 1 minute
Medium projects (20-100 files): 1-3 minutes
Large projects (100+ files): 3-5 minutes

Can I run verification on specific files?

Not directly, but you can:

Move to a subdirectory and run verification there
Use focused checks for faster feedback

What if verification finds nothing?

If you see ✅ All checks passed, your code follows best practices. The report will still contain the passing checks for documentation.

Does verification modify my code?

No. Agent Verifier only reads and analyzes—it never modifies files. All suggested fixes are recommendations for you to implement.

Next Steps

Now that you've run your first verification, learn how to interpret the results:

Tutorial 3: Understanding Reports →