Example Agents

Five example agents covering domains where reliability matters.

Each example in examples/ demonstrates a different agent domain with realistic stakes.

SQL Analyst

Domain: Finance/BI

A natural language to SQL agent backed by a SQLite database with customers and orders.

What can go wrong: Wrong revenue numbers, hallucinated metrics, incorrect joins/aggregations.

cd examples/sql-analyst
# In Claude Code: "evaluate this agent"

Incident Triage

Domain: SRE

An alert severity classification agent using LangGraph ReAct with a runbook knowledge base (OOMKilled, cert expiry, rate limits, disk usage, etc.).

What can go wrong: Missed P0, false 3am page, incorrect severity routing.

cd examples/incident-triage

Code Reviewer

Domain: Security

A code review agent checking for 10 vulnerability categories: SQL injection, secrets exposure, auth bypass, concurrency issues, error handling, and more.

What can go wrong: Shipped CVE, false positive fatigue, missed vulnerability in review.

cd examples/code-reviewer

Customer Support

Domain: CX

A ticket classification and response agent with a 10-article knowledge base and policy constraints (14-day refunds, GDPR 30-day deletion, enterprise escalation paths).

What can go wrong: Misrouted ticket, unauthorized refund promise, policy violation.

cd examples/customer-support

SDR Qualifier

Domain: Sales

A lead scoring and outreach agent with ICP segments, disqualification criteria, scoring rubrics, and outreach rules.

What can go wrong: Wasted AE time on unqualified leads, lost hot lead to slow follow-up.

cd examples/sdr-qualifier

Running an example

git clone https://github.com/satyaborg/kensa.git && cd kensa
uv sync --extra anthropic    # or --extra openai
cd examples/<agent-name>

Then in your coding agent:

> evaluate this agent