home / skills / getsentry / warden / agent-prompt

agent-prompt skill

safe

This skill helps you craft effective agent prompts and Warden skill prompts by applying reference guides and best practices.

npx playbooks add skill getsentry/warden --skill agent-prompt

Review the files below or copy the command above to add this skill to your agents.

Files (9)

SKILL.md

1.7 KB

---
name: agent-prompt
description: Reference guide for writing effective agent prompts and skills. Use when creating new skills, reviewing prompt quality, or understanding Warden's prompt architecture.
allowed-tools: Read Grep Glob
---

You are a prompt engineering specialist helping users write effective agent prompts and Warden skills.

## Reference Documents

The following documents contain detailed guidance. Read the relevant ones based on the user's question:

| Document | Use When |
|----------|----------|
| `references/core-principles.md` | Writing any prompt - foundational rules |
| `references/skill-structure.md` | Creating or reviewing skill files |
| `references/system-prompts.md` | Understanding Warden's prompt architecture |
| `references/output-formats.md` | Designing structured JSON output |
| `references/agentic-patterns.md` | Building tool-using agents |
| `references/anti-patterns.md` | Reviewing prompts for common mistakes |
| `references/model-guidance.md` | Optimizing for Claude 4.x models |
| `references/context-design.md` | Research on passive vs active context delivery |

## Quick Reference

**Skill file location**: `.agents/skills/{name}/SKILL.md`

**Minimum skill structure**:
```markdown
---
name: skill-name
description: One-line description for discovery.
allowed-tools: Read Grep Glob
---

[Role statement]

## Your Task

[What to analyze and criteria to apply]

## Severity Levels

[Definitions tied to impact]
```

## Your Task

When helping with prompts:

1. Read relevant reference documents before answering
2. Provide specific, actionable guidance
3. Show examples from existing Warden skills when helpful
4. Cite sources (Anthropic docs, etc.) for best practices

Overview

This skill is a practical reference for writing effective agent prompts and building Warden skills. It codifies prompt structure, evaluation criteria, and architecture guidance so creators can produce reliable, reviewable prompts. Use it to speed up authoring, review quality, and align prompts with Warden's operational model.

How this skill works

The skill inspects prompt structure, role statements, task descriptions, and severity definitions to ensure clarity and safety. It checks for common anti-patterns, enforces structured output formats, and recommends model-specific optimizations for Claude 4.x. It also maps prompts to agentic patterns when tool use or multi-step reasoning is required.

When to use it

Creating a new agent prompt or skill for Warden
Reviewing or auditing existing prompts for clarity and safety
Designing structured JSON outputs or integrating tools
Optimizing prompts for Claude 4.x or similar models
Training reviewers on prompt engineering best practices

Best practices

Start with a concise role statement and a clear single-sentence task objective
Define concrete acceptance criteria and severity levels tied to impact
Prefer structured, example-driven output schemas over free-form text
Avoid vague or open-ended instructions; constrain scope and expected steps
Explicitly separate system-level context from user-facing instructions
Test prompts with representative edge cases and refine using model feedback

Example use cases

Authoring a skill that inspects source code for security or style issues
Converting a manual checklist into an automated reviewer with tool calls
Creating a prompt that returns a validated JSON report for downstream tooling
Reviewing a prompt that chains tools and needs guardrails to prevent hallucination
Adapting a prompt to leverage Claude 4.x strengths like instruction-following and long context

FAQ

How do I structure severity levels?

Tie each severity to clear impact criteria and remediation steps so reviewers and agents can rank findings consistently.

Should outputs be free-form or structured?

Prefer structured JSON for machine consumption and include a short human summary when helpful; provide a schema and examples.