Skills and Automation

Skills are modular capabilities that you attach to agents and domain objects. They determine what an agent can do, when it activates, and what automations run on schedule. Understanding the three skill types unlocks the full power of Saltare's AI system.

The Three Skill Types

Tool Skills

Invoked through chat. When you @mention an agent, it selects from its available tool skills to handle your request.

Examples of tool skills:
- web_search — Search the internet
- create_task — Create a new task
- query_data_table — Query a database
- create_document — Create a new document

Tool skills are the most common type. By default, agents have access to all 69 built-in tools. You can restrict which tools an agent uses by configuring its skill set.

When to customize tool skills:
- You want a Researcher agent that can search the web but can't modify tasks
- You want a PM agent that manages tasks but can't access databases
- You want to limit an agent's scope to prevent unintended side effects

Trigger Skills

Activated by events. Instead of being invoked by a user, trigger skills fire automatically when something happens in the workspace.

Examples of trigger events:
- A new task is created
- A webhook payload arrives from GitHub
- A document is published
- A message mentions a specific keyword

How triggers work:
1. An event occurs (task created, webhook received, etc.)
2. The trigger skill's conditions are evaluated
3. If conditions match, the associated agent is activated
4. The agent processes the event and takes action

Example: Auto-triage GitHub issues
Set up a trigger skill on your PM agent:
- Event: Webhook from GitHub (new issue created)
- Action: Agent reads the issue, assigns a priority label, creates a matching task in Saltare, and posts a summary in #engineering

Automation Skills

Run on schedules. Automation skills use cron-like expressions to execute at regular intervals.

Examples:
- Daily standup summary — Every morning at 9 AM, summarize yesterday's activity
- Weekly report generation — Every Friday, compile metrics into a document
- Hourly monitoring — Check external APIs and alert on anomalies

Cron expression basics:

┌───── minute (0-59)
│ ┌───── hour (0-23)
│ │ ┌───── day of month (1-31)
│ │ │ ┌───── month (1-12)
│ │ │ │ ┌───── day of week (0-6, Sunday=0)
│ │ │ │ │
0 9 * * 1-5 → 9:00 AM, Monday through Friday
0 */4 * * * → Every 4 hours
30 17 * * 5 → 5:30 PM every Friday
0 8 1 * * → 8:00 AM on the 1st of each month


Combining Skill Types

The real power comes from combining all three types:

Example: Automated Content Pipeline

Tool skills (on Writer agent):
- create_document, update_document, web_search

Trigger skill (on Writer agent):
- When a task labeled "content-request" is created, automatically start drafting

Automation skill (on PM agent):
- Every Monday at 9 AM, check the content calendar database and create tasks for the week's planned content

Result: Content requests flow automatically from calendar → task → draft, with minimal manual intervention.

Example: Incident Response

Trigger skill (on Ops agent):
- When a Sentry webhook reports a critical error, create an urgent task and post in #engineering

Tool skills (on Ops agent):
- search_messages to find related discussions
- create_task to track the incident
- post_message to alert the team

Automation skill (on Ops agent):
- Every hour during business hours, check monitoring endpoints and report anomalies

Connector-Created Skills

When you set up a connector (e.g., GitHub, Slack), it can auto-create skills that grant agents access to the connector's tools. For example:

• Setting up a GitHub connector creates skills for github_list_issues, github_create_issue, etc.
• Setting up a Google Calendar connector creates skills for google_calendar_list_events, google_calendar_create_event, etc.

These connector-created skills automatically appear in the agent's available tools.

Managing Skills

Viewing Skills

Each agent's profile shows its attached skills. You can see which tools, triggers, and automations are active.

Adding Skills

Attach new skills to an agent from the agent's settings page. Choose the skill type, configure its parameters, and activate it.

Removing Skills

Remove skills to restrict an agent's capabilities. The agent will no longer have access to the removed tools or respond to the removed triggers.

Best Practices

• Start with defaults — New agents get all tool skills. Only restrict when you have a reason.
• Use triggers sparingly — Too many triggers create noise. Start with one or two high-value automations.
• Test automations in off-hours — Schedule initial test runs at times when you can monitor the output.
• Monitor skill usage — Check which skills agents actually use to identify unnecessary permissions.
• Document your automations — Keep a list of active trigger and automation skills so your team knows what's running automatically.