Prompt Engineering Quick Reference

Prompt Engineering Core Patterns

1. Role Specification

State the role clearly at the top of the prompt.

You are a senior security analyst reviewing endpoint and identity evidence for incident triage.

Why it helps: role framing sets expected depth, vocabulary, and decision style.

2. Output Contract

Specify exact output structure.

Output as a markdown table with columns:
Control ID | Coverage Level | Compliance Status | Rationale

Why it helps: strict schemas make output predictable and easier to review or automate.

3. Constraints and Decision Rules

Define boundaries and tie-break behavior.

If evidence suggests multiple categories, choose the highest-risk category and explain why.
Only include Critical and High severity findings in the remediation list.

Why it helps: constraints reduce drift and force deterministic prioritization.

4. Confidence Labels

Require confidence per finding.

Confidence levels:
- HIGH: clear direct evidence
- MEDIUM: likely but incomplete evidence
- LOW: ambiguous or insufficient evidence

Why it helps: confidence makes recommendations operationally actionable.

Start simple, run, inspect, and refine.

Iteration 1: basic request
Iteration 2: role + schema
Iteration 3: categories + confidence + edge-case handling

Why it helps: real prompts improve through controlled iteration, not one-shot drafting.

Knowledge Store Tips

Attach only files relevant to the current task.
Prefer structured formats (Markdown, CSV, JSON) when possible.
Keep references current and remove stale material.
Split large documents into focused sections to reduce retrieval noise.

Reusable Prompt Checklist

Use this checklist before running an agent:

Role is explicit
Objective is explicit
Required categories or labels are explicit
Output schema is explicit
Confidence rules are explicit
Edge-case handling is explicit
Post-processing instructions are explicit (summary, priorities, next steps)

Kindo Documentation Links

Agent Construction Best Practices