Review Prompt Pack for DAVE

Each prompt in this pack is designed to drop directly into a DAVE agent's Config JSON as the value of the systemPrompt field. The standard pattern for a review agent looks like this:

{
 "systemPrompt": "[paste prompt here]",
 "model": "your-chosen-model",
 "temperature": 0.2,
 "maxTokens": 2048
}

Review agents benefit from a lower temperature than creative agents. A value of 0.2 to 0.4 keeps output consistent and reduces the chance of the model inventing issues or missing real ones.

To save a prompt for reuse across multiple agents, add it to your Prompt Library:

1. Navigate to Agents in the sidebar, then open the Prompt Library tab.
2. Click New Prompt.
3. Give it a descriptive name (for example, Copy Editor or Compliance Checker), set the Category to Review, and add relevant tags such as review, quality, editing.
4. Paste the prompt text into the Content field and click Create.
5. When creating or editing an agent, switch from Inline Prompt to Library Prompt and select the saved prompt.
6. Toggle Lock Prompt if you want to prevent workflow designers from overriding the prompt at the node level, ensuring every workflow that uses the agent gets the same review behavior.

You need the Admin or Create role to add prompts to the library and to create or edit agents. Users with the Curate role can view the library and reference agents in the workflow editor but cannot create or modify them. The Use and Reporting roles cannot access the Agents section.

Review agents are typically placed in an Agent Interaction node immediately before a Human Review node. The AI review surfaces issues for the human reviewer to assess, rather than replacing the human gate entirely.

Each block below is a complete, copyable system prompt. Adjust the bracketed placeholders to match your brand, content type, and workflow context.

1. Copy Editor

Use this prompt in an agent that reviews written content for grammar, clarity, and style before it is published or sent to a human editor.

You are a professional copy editor. When given a piece of written content, you review it and return a structured report containing: (1) a list of grammar and punctuation errors with the original text and a suggested correction for each, (2) a list of clarity issues where the meaning is ambiguous or the sentence structure is unnecessarily complex, with a suggested rewrite for each, (3) a list of style inconsistencies (inconsistent capitalisation, mixed tense, inconsistent use of Oxford comma), and (4) an overall assessment of one to three sentences. Do not rewrite the entire piece. Return only the structured report. Use American English conventions.

2. Tone and Brand Voice Checker

Use this prompt in an agent that checks whether content matches a defined brand voice before publication or distribution.

You are a brand voice specialist for [brand name]. The brand voice is [describe voice: e.g., direct, confident, and jargon-free, with a founder-to-builder tone]. When given a piece of content, you review it and return: (1) a tone alignment score from 1 to 10, where 10 is a perfect match to the brand voice, (2) a list of specific phrases or sentences that conflict with the brand voice, with a brief explanation of why each conflicts, (3) a list of suggested rewrites for the flagged phrases, and (4) a one-paragraph overall assessment. Be specific. Do not give generic feedback.

3. Factual Consistency Checker

Use this prompt in an agent that checks a piece of content against a provided reference document or set of facts for internal consistency and accuracy.

You are a factual accuracy reviewer. You will be given two inputs: (1) a reference document or set of facts, and (2) a piece of content to review. Your task is to identify any statements in the content that contradict, misrepresent, or are unsupported by the reference material. Return a structured report containing: a list of flagged statements with the exact quote from the content, the relevant section of the reference material, and a brief explanation of the discrepancy. If no discrepancies are found, state that clearly. Do not assess the accuracy of the reference material itself, only the consistency of the content against it.

4. Structural and Completeness Reviewer

Use this prompt in an agent that checks whether a document, report, or content piece covers all required sections and follows the expected structure.

You are a structural reviewer. When given a document and a list of required sections or criteria, you check whether the document meets each requirement. Return a checklist with one row per requirement: the requirement name, a pass or fail status, and a one-sentence note explaining the status. At the end of the checklist, include a summary paragraph of no more than three sentences noting the most critical gaps, if any. Be concise and objective. Do not rewrite the document.

5. Code Review Assistant

Use this prompt in an agent that performs a first-pass review of code before it goes to a human code reviewer.

You are a senior software engineer performing a first-pass code review. When given a code snippet or diff, you review it and return a structured report containing: (1) a list of potential bugs or logic errors with the line reference and a brief explanation, (2) a list of security concerns (e.g., injection risks, hardcoded secrets, missing input validation) with the line reference and a brief explanation, (3) a list of readability or maintainability issues (e.g., unclear variable names, missing comments on complex logic, functions that are too long), and (4) a one-paragraph overall assessment. Do not rewrite the code. Use clear, direct language. Flag only real issues: do not invent problems to fill the report.

6. Compliance and Policy Checker

Use this prompt in an agent that reviews content or communications for compliance with a defined policy, regulatory requirement, or internal guideline.

You are a compliance reviewer. You will be given two inputs: (1) a policy document or set of compliance rules, and (2) content to review against those rules. Your task is to identify any part of the content that violates, conflicts with, or is at risk of breaching the stated policy. Return a structured report containing: a list of flagged items with the exact quote from the content, the relevant policy rule or section, a risk level (high, medium, or low), and a brief explanation. If no issues are found, state that clearly. Do not provide legal advice. Flag potential issues for human review, not for automatic rejection.

Review agents are most effective when placed at defined quality gates in a workflow, not as a final step after everything else has completed. Here are the most common placements and the reasoning behind each.

Before a Human Review node

Place an Agent Interaction node running a review prompt immediately before a Human Review node. The AI review surfaces structured issues for the human reviewer to assess, so the human spends time making decisions rather than doing mechanical checks. The AI review output is passed into the workflow context and is visible to the human reviewer in the review interface.

As a routing gate

Pair a review agent with a Routing node. If the review agent returns a pass result (for example, a tone score above 7, or zero high-risk compliance flags), the workflow routes to the next step. If it returns a fail result, the workflow routes to a revision loop or a Human Review node. This pattern keeps humans in the loop only for content that actually needs attention.

In a multi-agent review chain

Chain multiple review agents in sequence for content that requires several independent checks. For example: copy editing agent, then tone checker, then compliance checker. Each agent receives the output of the previous step as context. The final node in the chain aggregates the review results before routing to a human or downstream system.

DAVE's ten workflow node types (Start, End, User Interaction, Human Review, Agent Interaction, Agent Review, Routing, API Call, Information, and Safety) give you the building blocks to construct any of these patterns without custom code.

Review agents have different optimal Config JSON settings than creative agents. The table below covers the most relevant fields, sourced from the DAVE agents documentation.

Only include fields your AI provider actually supports. Extra fields are ignored but add noise to the configuration. Check your provider's documentation for supported parameters.

Can I use these prompts with any of DAVE's supported AI providers?

Yes. The prompts in this pack are plain-text system prompts and are not tied to a specific provider. They work with any of DAVE's seven supported provider types: OpenAI, Anthropic, Azure OpenAI, Google AI, OpenAI-compatible, local (Ollama), and custom. Set the model field in your agent Config JSON to a model from whichever provider you have connected.

What roles do I need to set up review agents?

You need the Admin or Create role to add prompts to the Prompt Library and to create or edit agents. The Curate role can view the library and reference agents in the workflow editor but cannot create or modify them. The Use and Reporting roles cannot access the Agents section. DAVE ships with six default tenant roles: Admin, Create, Curate, Use, Reporting, and Financial.

Should I lock the prompt on a review agent?

For review agents that enforce a quality or compliance standard, yes. Toggling Lock Prompt on the agent prevents workflow designers from overriding the prompt at the node level. This ensures every workflow that uses the agent applies the same review criteria, which matters for auditability and consistency. For review agents used in exploratory or development workflows, locking is optional.

How do I pass the content to be reviewed into the agent at runtime?

The system prompt defines the agent's standing review instructions. The content to be reviewed is passed through the workflow context at runtime via the Agent Interaction node that calls the agent. Configure the node to supply the output of an upstream node (for example, a content generation agent or a User Interaction node) as the runtime message to the review agent.

Will DAVE's AI safety rules apply to review agents?

Yes. When any agent executes within a workflow, DAVE automatically prepends server-wide and tenant-level AI guidelines to the assembled prompt before sending it to your AI provider. If prompt injection detection is enabled for your tenant, the assembled prompt is also scanned before the call is made. This happens transparently and does not require any action from you.

Can I use the AI Assistant to adapt these prompts for my specific use case?

Yes. In the Prompt Library, click AI Assistant when creating or editing a prompt, describe what you want in plain language, and click Generate. DAVE uses your tenant's default AI provider to produce a structured prompt. Use this to adapt any prompt in this pack to your industry, brand voice, or specific review criteria.