Actions

Actions define what your automation does when a trigger fires. Currently, Cheffed supports the Start Session action.

Start Session Action

Creates a new AI session when the automation triggers.

Configuration Options

Option	Description	Required
Agent/Prompt	Which agent template to use	Yes
Model	AI model to power the session	Yes
Repository	Clone a repo into the session	Optional
Initial Message	First message sent to the agent	Optional

Agent Selection

Choose an agent that matches your automation’s purpose:

Use Case	Recommended Agent
Bug triage	Bug Analyzer
Code review	Code Reviewer
Feature implementation	Feature Builder
Documentation	Doc Generator
Custom workflow	Your custom agent

Model Selection

Select the AI model for the session:

Model	Best For
Claude 3.5 Sonnet	Balance of speed and capability
Claude 3 Opus	Complex reasoning tasks
GPT-4	Broad knowledge, general tasks
GPT-4 Turbo	Faster GPT-4 responses

Repository Configuration

Connect a repository to the session:

Same as trigger - Use the repo from the triggering event
Specific repo - Always clone a particular repository
None - No repository cloned

When using “Same as trigger,” the repository from the GitHub event is automatically cloned.

Initial Message

Craft the first message the agent receives. This can include:

Static text:

Analyze this issue and provide initial triage findings.

Dynamic placeholders:

Please analyze issue #{{issue.number}}: {{issue.title}}

Issue body:
{{issue.body}}

Provide:
1. Root cause analysis
2. Affected components
3. Suggested fix approach

Available Placeholders

For GitHub Issue triggers:

Placeholder	Description
`{{issue.number}}`	Issue number
`{{issue.title}}`	Issue title
`{{issue.body}}`	Issue body/description
`{{issue.author}}`	Issue author username
`{{issue.labels}}`	Comma-separated label names
`{{issue.url}}`	Full URL to the issue

For GitHub PR triggers:

Placeholder	Description
`{{pr.number}}`	PR number
`{{pr.title}}`	PR title
`{{pr.body}}`	PR description
`{{pr.author}}`	PR author username
`{{pr.head_branch}}`	Source branch
`{{pr.base_branch}}`	Target branch
`{{pr.url}}`	Full URL to the PR

For all triggers:

Placeholder	Description
`{{repository.name}}`	Repository name
`{{repository.owner}}`	Repository owner
`{{repository.full_name}}`	`owner/name` format

Action Behavior

Session Lifecycle

When an action creates a session:

Container provisioned - Isolated environment spins up
Repository cloned - If configured
Agent initialized - Prompt template loaded
Initial message sent - Your configured message
Agent works - Processes the request
Results produced - Code changes, comments, reports

Output Handling

Session results can include:

GitHub comments - Agent posts findings to the issue/PR
Pull requests - Agent creates a PR with changes
Reports - Analysis saved as session artifacts

Error Handling

If an action fails:

Error logged in execution history
Session marked as failed
No partial results applied
Retry available from execution history

Advanced Configuration

Timeout Settings

Configure how long the action can run:

Default: 30 minutes
Maximum: 2 hours

Sessions that exceed timeout are terminated gracefully.

Conditional Actions

Coming soon: Run different actions based on trigger content.

Best Practices

Craft Clear Initial Messages

The initial message sets the agent’s direction. Make it:

Specific - Clear task description
Contextual - Include relevant trigger data
Actionable - Define expected outputs

Match Agent to Task

Choose agents designed for your automation’s purpose. A bug analyzer agent will perform better on bug triage than a generic coding agent.

Test Before Deploying

Create the automation
Manually trigger with a test event
Review the session output
Adjust configuration as needed

Next Steps

Build your first automation
Configure event triggers
Monitor execution history