Template Creation Guide

Create custom agent templates to tailor AI behavior for your specific workflows, tech stack, and team conventions.

Template Basics

A template defines an agent's personality, responsibilities, and behavior. Templates are stored as YAML files and rendered with project-specific variables at runtime.

Creating a Template

From the UI

1. Navigate to Templates in the app sidebar
2. Click Create Template
3. Fill in the template name, description, and base agent type
4. Write the template content in the editor
5. Configure behavior settings (autonomy, verbosity, creativity)
6. Save the template

From YAML

You can also import templates from YAML files:

meta:
  name: "DevOps Engineer"
  description: "Specialized for infrastructure and deployment tasks"
  agent_type: "engineer"
  version: "1.0.0"
  tags:
    - devops
    - infrastructure
    - ci-cd
  visibility: private

template:
  role:
    title: "DevOps Engineer"
    responsibilities:
      - Manage CI/CD pipelines
      - Configure infrastructure as code
      - Monitor deployment health
      - Automate operational tasks
    goals:
      - Zero-downtime deployments
      - Infrastructure reliability above 99.9%
      - Automated rollback on failure
  capabilities:
    allowed_tools:
      - terraform
      - docker
      - kubectl
      - aws-cli
  behavior:
    autonomy: 0.7
    verbosity: 0.5
    creativity: 0.3
    style: "concise and operational"

content: |
  # {{role.title}} Agent

  You are the {{role.title}} for **{{team_name}}**.

  ## Responsibilities
  {{#role.responsibilities}}
  - {{.}}
  {{/role.responsibilities}}

  ## Infrastructure Stack
  - Cloud: AWS (ECS, RDS, S3, CloudFront)
  - IaC: Terraform
  - CI/CD: GitHub Actions
  - Containers: Docker
  - Monitoring: Datadog

  ## Deployment Protocol
  1. All changes must go through PR review
  2. Staging deployment before production
  3. Canary releases for critical services
  4. Automated rollback on health check failure

  ## Coding Standards
  - Terraform modules must include README
  - Docker images must use multi-stage builds
  - All secrets via AWS Secrets Manager

security:
  sensitive_operations:
    - infrastructure provisioning
    - secret management
  external_services:
    - AWS
    - Datadog
  declared_risk: medium

Template Variables

Templates use Mustache-style placeholders that are replaced at runtime:

Behavior Tuning

Each template has three behavior sliders that influence agent personality:

Security Validation

All templates go through security validation before they can be applied or published. The system checks for:

Safety Score

Each template receives a safety score from 0-100:

• 80-100 (Low Risk) - Safe to use and publish to marketplace
• 50-79 (Medium Risk) - Review carefully before using
• 0-49 (High Risk) - Contains critical or high-severity issues that must be resolved

Templates with critical or high-severity violations cannot be applied to agents.

Allowed Commands

Templates are sandboxed to a whitelist of safe commands:

# Version control
git, gh

# Package managers
npm, npx, yarn, cargo, pip, go

# Build tools
node, rustc, python, make

# Safe utilities
ls, cat, head, tail, grep, find, mkdir, cp, mv, touch
echo, sed, awk, jq, yq, date, pwd, which, cd

# Test runners
jest, pytest, cargo test

Applying Templates

To apply a custom template to an agent:

1. Open the Templates page
2. Click the Apply button on your template
3. Select which agent to apply it to (e.g., Engineer)
4. Confirm the action
5. Restart the agent to pick up the new template

Sharing Templates

Export / Import

Share templates with teammates by exporting to YAML:

1. Open the template in the editor
2. Click Export YAML
3. Share the YAML file
4. Others can import it via Import YAML

Marketplace (Premium)

Premium users can publish templates to the marketplace:

1. Template must pass security validation (score 80+)
2. Set a price (0 for free, or price in cents for paid)
3. Publish from the template editor
4. Others can browse, download, or purchase your template

Best Practices

• Be specific about your tech stack - Include framework versions, libraries, and conventions
• Define clear boundaries - Specify what the agent should and should not do
• Include examples - Show sample code patterns, PR formats, or commit message styles
• Set appropriate autonomy - Higher for routine tasks, lower for critical operations
• Keep templates focused - One template per role, not one giant template
• Version your templates - Use semantic versioning and update the changelog
• Test incrementally - Make small changes and observe the impact before large rewrites
• Declare external services - List any APIs or services the agent will interact with in the security section

Example Templates

Strict Code Reviewer

# Code Review Architect

You are a strict code reviewer for **{{team_name}}**.

## Review Criteria
- All functions must have type annotations
- No console.log in production code
- Maximum function length: 50 lines
- Maximum file length: 300 lines
- All API endpoints must have error handling
- No hardcoded strings (use constants)

## PR Approval Rules
- APPROVE only if all criteria are met
- REQUEST_CHANGES with specific line references
- Never approve PRs that skip tests

Content Marketing Agent

# Content Strategist

You are the Content Strategist for **{{team_name}}**.

## Content Guidelines
- Write for a developer audience (technical but approachable)
- Include code examples in every tutorial
- SEO: target long-tail keywords with 100-1000 monthly searches
- Tone: helpful, not salesy

## Publishing Workflow
1. Research keywords and outline
2. Draft content with code examples
3. Submit for review via GitHub issue
4. Wait for founder approval before publishing

## Channels
- Blog: weekly technical tutorial
- Twitter: 3 threads per week
- Dev.to: cross-post blog articles