Effective Prompting with Harness AI
Harness AI generates pipelines, services, environments, connectors, and secrets from natural language.
This guide shows you how to write effective prompts with examples of weak vs. strong phrasing.
Why Prompting Matters
Harness AI performs best when prompts are clear, detailed, and specific.
The difference between “create a pipeline” and “create a CI→CD pipeline that builds a Docker image and deploys to Kubernetes” is the difference between vague YAML and production-ready YAML.
Prompting Guidelines
- Be specific: Name repos, branches, connectors, and strategies.
- Use action words: Start with Generate, Create, Define.
- Provide inputs: List parameters like
IMAGE_TAG,REPLICA_COUNT. - Focus: One resource per prompt is more reliable.
- Iterate: Begin broad, refine with more detail.
Pipeline Prompts
| Weak Prompt | Strong Prompt | Why It’s Better |
|---|---|---|
Create a pipeline for my app | Generate a Harness CI→CD pipeline that builds a Docker image and pushes to Dockerhub. Add a CD stage to deploy a Kubernetes service to namespace dev on cluster co[...] | More detailed, includes build, push, and deployment targets. |
Make a pipeline | Create a pipeline that runs unit tests for python, builds with Dockerfile, scans image with semgrep and deploys to staging using a Rolling strategy. Use GitHub connector rohan-git. | Specifies language, build, scan, deploy, strategy, and connector. |
Service Prompts
| Weak Prompt | Strong Prompt | Why It’s Better |
|---|---|---|
Make a Kubernetes service | Generate a Kubernetes service definition named portal using Helm with chart path /cd/chart/portal. Expose variables: IMAGE, TAG, REPLICA_COUNT. Include artifact references. | Names, chart path, exposed variables, artifact refs. |
Create a Lambda service | Create a serverless (AWS Lambda) service named router with runtime nodejs18.x, artifact from runtime input, and environment variables LOG_LEVEL, STAGE. | Adds runtime, artifact source, and environment variables. |
Environment Prompts
| Weak Prompt | Strong Prompt | Why It’s Better |
|---|---|---|
Create an environment | Generate an Environment named staging (type: PreProduction) | Defines name and environment type explicitly. |
Make prod environment | Create a Production environment named prod | Clear scope for production use. |
Secret Prompts
Important: Harness creates the secret object without values. You must add values manually in the UI. Secret values are never sent to AI.
| Weak Prompt | Strong Prompt | Why It’s Better |
|---|---|---|
Add a secret | Create a Secret Text named docker-hub-token. Store in default Secret Manager. | Includes type, name, and storage location. |
Connector Prompts
| Weak Prompt | Strong Prompt | Why It’s Better |
|---|---|---|
Make a Docker connector | Generate a Docker Registry connector named canary-docker pointing to harness/canary with credentials from secret refs usernameRef: canary and passwordRef: docker-hub-token. | Names, repo, and credential refs. |
Create a Vault connector | Create a HashiCorp Vault connector, called dev-vault and default the rest of the values. | Identifies type, name, and defaults. |
Template Prompts
| Weak Prompt | Strong Prompt | Why It’s Better |
|---|---|---|
Use a template | Create a Pipeline with the Deploy to K8s Stage Template. | Directly references a stage template by name. |
Add template | Create a Pipeline with the Golden K8s Pipeline Template. | Specifies which pipeline template to reuse. |
Additional Prompting Guidelines for Harness AI
Beyond the basics, here are more advanced prompting techniques to get consistent and high-quality results from Harness AI.
Other Prompting Guidelines
1. State the Resource Scope
Always specify whether the resource should be created at the Account, Org, or Project level.
Example
Generate a pipeline at the Project scope that uses the staging environment and GitHub connector rohan-git.
2. Include Runtime Inputs
Call out variables you want parameterized with <+input>.
Example
Create a Kubernetes service with IMAGE and TAG as runtime inputs, defaulting to `<+input>`.
3. Specify Deployment Strategies
Explicitly mention strategies such as Rolling, Blue/Green, or Canary.
Example
Create a production pipeline with a Canary deployment using 25%-75% rollout strategy.
4. Chain Stages Clearly
Describe the order of stages and include gating conditions.
Example
Generate a CI pipeline with build → test → deploy stages,
add a manual approval before deploy,
and notify Slack after success.
6. Add Verification or Notifications
Ask AI to extend pipelines with observability or collaboration integrations.
Example
Extend the staging deployment pipeline to run a health check verification step
and post results to Slack channel #deployments.
7. Use Iterative Prompting
Start with a basic resource and refine by adding details in follow-up prompts.
Example
First Prompt
Create a pipeline that builds and deploys to staging.
Follow up Prompt
Add a canary rollout strategy to the deployment stage.
Best Practice Checklist
- Mention target platform (Kubernetes, ECS, Lambda, VM)
- Include connector names and secret references
- Add environment/namespace details
- Specify deployment strategy (Rolling, Canary, Blue/Green)
- Use variables for flexibility (
IMAGE_TAG,SERVICE_NAME)
Next Steps
- Start with the strong prompt examples in the tables above.
- Adjust connectors, namespaces, and service names to match your project.
- Run, review, and iterate to accelerate CI/CD setup with Harness AI.