Paige: Connectors and Skills

Paige, PagerDuty's SRE Agent, retrieves log data, metrics, and runbooks from observability platforms and knowledge bases to guide responders through investigation, triage, and resolution. By connecting your observability stack and institutional knowledge, Paige moves beyond simple alert analysis.

• Connectors and Tools give Paige real-time access to your logs and metrics, so every triage analysis is grounded in the actual state of your systems — not just the alert payload.
• Skills give Paige institutional knowledge specific to your environment — your runbooks, your escalation paths, your standards — so it can recommend precise next steps instead of generic ones.

Together, these capabilities help your team triage faster with less human involvement, reducing MTTA and MTTR while building confidence in AI-assisted operations.

To get started, navigate to Configuration AI AI Settings and click Configure on the Paige card under AI Agents. Paige Configuration opens.

Connectors and Tools

Connectors link Paige to your external data sources. Each connector provides a set of tools — individual actions such as querying logs, reading a file, or listing pull requests. You choose which tools Paige may use for each connector, and Paige uses only the allowed tools, automatically or via user request.

Supported Connectors

Navigate to Configuration AI AI Settings, then click Configure on the Paige card under AI Agents to open Paige Configuration, and select the Connectors tab to view and manage your connectors.

Each tool category contains individual tools that you can allow or block. For example, GitHub's Code category includes tools such as Search Code, List Pull Requests, Get File Contents, and List Commits.

Set up a Connector

1. In the PagerDuty web app, navigate to Configuration AI AI Settings, then click Configure on the Paige card under AI Agents to open Paige Configuration, and click the Connectors tab.
2. Find the connector you want to add and click Connect. If the connector is already listed but shows Unhealthy, click the menu to edit or reconnect it.
3. On the Connect to [Integration] Tools screen, select an existing connection from the dropdown or click Manage Connections to create a new one.
4. Choose which tools Paige may use. Tools are grouped into Read Only Tools and Write/delete Tools; set each tool, or a whole group, to Always Allow or Blocked.
5. Click Save. The connector appears in your list with a status of Active once the connection is established.

Connector Statuses

• Active: The connector is healthy and Paige can use its tools.
• Unhealthy: The connection exists but is failing. Check your credentials or API key and reconnect.
• Not Connected: No connection has been set up. Click Connect to configure it.

Connector Setup Notes

Atlassian

The Atlassian connector lets Paige search Confluence and read pages through Atlassian's remote MCP server. It uses OAuth 2.1 with PKCE and doesn't need a client secret or API key. Paige's access is read-only: it can search Confluence and read pages, and it can't create or change content.

To connect Atlassian, follow Set up a Connector and select Atlassian. When you create a new connection, you sign in to Atlassian, select the site to integrate, and approve access.

PagerDuty requests read access to Confluence content, Confluence search, Rovo search, and your basic Atlassian profile. Paige uses these to list the Atlassian sites you can access, search Confluence in natural language or with CQL, and read a Confluence page by ID or URL.

Searches that Paige runs through the Atlassian connector use Rovo credits from your Atlassian plan.

Atlassian access tokens last 8 hours. PagerDuty checks the connection and refreshes the token every hour. If the connector shows Unhealthy, reconnect it to sign in to Atlassian again.

Confluence

The Confluence connector supports semantic search: Paige can search your Confluence knowledge base using natural-language queries, and results are matched to the meaning of the request rather than requiring exact keyword matches.

Notion

The Notion connector connects Paige to a Notion workspace through Notion's hosted MCP server. It uses OAuth 2.1 with PKCE and doesn't need a client secret or API key. Paige's access is read-only: it can search and read pages, databases, and data sources, and it can't create, edit, or delete Notion content. You can also paste a Notion page URL in chat so Paige reads that page.

To connect Notion, follow Set up a Connector and select Notion. When you create a new connection, you sign in to Notion and approve access.

Each connection covers one Notion workspace: the one you select when you approve access.

When the workspace has a Notion AI plan, Paige uses Notion's AI (natural-language) search. Otherwise, Paige falls back to keyword search. Paige doesn't show Notion upgrade links in either case.

Notion access tokens last 8 hours, and the connection health check refreshes them hourly. Notion requires you to re-authorize after 180 days, or after 30 days without use. If the connector shows as Unhealthy, reconnect it to fix it.

Splunk

The Splunk Metrics and Traces connectors use the Splunk MCP server (/services/mcp) and reuse your existing Splunk connection. The MCP server accepts only tokens created with the mcp audience (audience=mcp) and rejects other tokens with a 403 error.

The Traces connector reads spans indexed as events in Splunk Enterprise and Splunk Cloud Platform. Splunk Observability Cloud APM is not supported.

Runbook Links

When Paige is provided access to documents, it fetches a document if it determines the document is relevant or if you request it in chat.

Confluence

1. Set up the Workflow Integration and set up the connector. Alternatively, set up the Atlassian connector, which doesn't need a Workflow Integration. Paige fetches a Confluence runbook link through whichever one is enabled for your account.

2. If you have your runbook link configured, Paige should identify it in the event payload.
   1. If Paige doesn't identify the runbook link, update the event payload in your monitoring tool to include your Confluence runbook link in custom_details using the following nested key:

"custom_details": {
    "runbook_url": {
        "confluence": "https://YOUR-RUNBOOK-LINK"
    }
}

GitHub

1. Set up the GitHub Cloud connector and set up the connector.

2. If you have your runbook link configured, Paige should identify it in the event payload.
   1. If Paige doesn't identify the runbook link, update the event payload in your monitoring tool to include your GitHub runbook link in custom_details using the following nested key:

"custom_details": {
    "runbook_url": {
        "github": "https://YOUR-RUNBOOK-LINK"
    }
}

ServiceNow

1. Set up the Workflow Integration and set up the connector.
2. If you have your runbook link configured, Paige should identify it in the event payload.
   1. If Paige doesn't identify the runbook link, update the event payload in your monitoring tool to include your ServiceNow runbook link in custom_details using the following nested key:

"custom_details": {
    "runbook_url": {
        "servicenow": "https://YOUR-RUNBOOK-LINK"
    }
}

Set Custom Details with Orchestration Rules

You can use orchestration rules to update the event payload to include your runbook links in custom_details.

1. Navigate to an orchestration.
2. Navigate to Edit Event Rule Event Fields Add Event Field.
3. Click the Select Field dropdown and add a custom details path to the event field:
   • For GitHub: runbook_url.github
   • For Confluence: runbook_url.confluence
   • For ServiceNow: runbook_url.servicenow

4. Under Replace using, select Template.
5. In Replace with value, paste the URL of the runbook.
6. Click Save.

Best Practices

• Log queries work best when there are query examples in the runbook or alert payload. Sample queries in the alert payload can be in any custom_details field.
• You can also provide queries directly to the agent to help ensure accurate log search.

Prompt Examples

• @pagerduty can you fetch the logs for this incident?
• @pagerduty can you search the logs with this criteria [enter criteria]?

Skills

Skills let you give Paige custom instructions, runbooks, and scripts for specialized domain expertise. Rather than relying only on retrieved data, a skill tells the agent how to handle specific situations, so it can work out the next steps for troubleshooting an issue.

How Paige Uses Skills

Skills load progressively to keep Paige's context focused. Paige does not read every skill's full instructions up front.

Paige may load a skill automatically if it decides the skill is relevant. After loading it, Paige may choose not to run the skill if the instructions do not fit the problem it is working on.

What a Skill Contains

Each skill is a structured set of instructions with the following components:

Example Skill: detect-security-incident-logs

The following is an example of a fully structured skill that you can use as a template. This skill searches Datadog and AWS CloudWatch logs for critical security events.

Name: detect-security-incident-logs

Description: Searches Datadog and CloudWatch logs for critical security events. Use when an incident may involve a security issue or a responder asks to check for security incidents.

Example Instructions:

When to use:
- The incident or alert suggests a possible security issue (for example, alert custom details include type: possible_security_issue)
- A responder asks the agent to check for security incidents

Prerequisites:
- Datadog connector configured and active
- AWS CloudWatch connector configured and active

Execution steps:
1. Search Datadog logs for the exact phrase SECURITY_INCIDENT: CRITICAL within the incident time window.
2. Search AWS CloudWatch logs for the exact phrase SECURITY_INCIDENT: CRITICAL within the incident time window.
3. If matches are found in either source, surface the key log details in Paige's response.
4. Recommend triggering the Alert Security Team incident workflow.

Error handling:
- If one connector is unavailable, continue with the other and note which source was skipped.
- If both connectors are unavailable, inform the responder and suggest checking logs manually.

Success criteria:
- The agent has queried both log sources and reported results or confirmed no matches were found.

Examples:
- "Check for any security incidents in the logs"
- "Are there critical security events related to this incident?"
- Triaging an alert where custom_details includes type: possible_security_issue

Create a Skill

You can create skills in the PagerDuty web app or with the PagerDuty Claude plugin.

In the PagerDuty Web App

1. In the PagerDuty web app, navigate to Configuration AI AI Settings, then click Configure on the Paige card under AI Agents to open Paige Configuration, and click the Skills tab.
2. Click Add Skill.
3. On the New Skill page, complete the following fields:

4. Click Save. Paige can use the skill on future incidents.

Use the Claude Plugin

1. Install the PagerDuty Claude plugin.
2. Run the /pagerduty:create-pagerduty-skill command and follow the guided prompts to define the skill's name, description, instructions, usage guidance, and examples.
3. Once created, Paige can use the skill on future incidents.

Test a Skill

After creating a skill, test it by asking Paige to use it in chat on a relevant incident, or by triggering an incident that your runbook or triage skill links to the skill. Then review what the agent reports in:

• The Paige tab on the incident details page.
• Your ChatOps channel (Slack or Microsoft Teams).