SoftTech AI Registry
GitHub · GitHub Copilot

Copilot Automations

Persistent, user-configured triggers and tool permissions that start GitHub Copilot cloud agent sessions automatically on a schedule or in response to repository events, without a person initiating each individual run.

Recorded characteristics

Function
GitHub documents Copilot automations as a way to run Copilot cloud agent automatically, on a schedule or in response to events in a repository. A user with write access defines an automation once, consisting of a name, a prompt describing the task, one or more triggers, the model Copilot uses, and the tools Copilot can use. When a trigger fires, GitHub starts a Copilot cloud agent session that runs the configured prompt with the selected tools, without a person initiating that individual run. Documented example uses include triaging incoming issues by labelling them, checking for failing tests on the main branch nightly and attempting a fix with a draft pull request, and drafting weekly release notes via a pull request. An automation can only take action in the single repository it is scoped to. Automations can also be run immediately from the Automations pane using a Run now button for testing. Automations are distinct from a manually started Copilot cloud agent session, which requires the user to give Copilot a task each time; the automation supplies the persistent trigger, prompt and tool configuration, while execution itself is performed by Copilot cloud agent. GitHub separately documents GitHub Agentic Workflows for storing automation as code, reviewing it through pull requests, or running it with a different coding agent.
Data access
GitHub documents that automations use the Copilot cloud agent configuration for the repository they are scoped to, including repository custom instructions, agent skills, firewall rules, and secrets and variables configured for Copilot cloud agent. Inputs also include the automation prompt written by its creator and the repository event that fired the trigger; event-based triggers can be narrowed with a search query filter, and pull request triggers additionally with a files-changed filter. Automations are scoped to a single repository and are stored separately from repository contents, so they are not committed to Git. GitHub warns that sessions started by an automation, including the prompt and session logs, are visible to people with repository access, and advises using repository secrets rather than placing sensitive values in the prompt. Access beyond the documented repository scope and inherited cloud agent configuration is not established.
Actions
Can take actions
External actions
Conditional
Human confirmation
Conditional
Permission basis
Mixed
Administrative control
GitHub documents that organizations can control whether automations are allowed in a repository separately from the Copilot cloud agent policy, using an "Allow automations" toggle under the organization Copilot cloud agent settings; automations are allowed by default in repositories where Copilot cloud agent is available. Automations are also unavailable where a repository administrator has disabled them, in repositories owned by managed user accounts, in public repositories, and where Copilot cloud agent is not enabled. For Copilot Business and Copilot Enterprise, an administrator must first enable the Copilot cloud agent policy, and organization owners can restrict which repositories allow the cloud agent. Repository-level automation controls include the tools selected for each automation, event-trigger filters, and the default behaviour of ignoring events triggered by users without write access. Repositories also have an automation level (Full control, Cautious as default, Balanced, Full automation) that sets the confidence threshold above which issue changes apply automatically; GitHub states these approvals are a workflow convenience rather than a security control. Firewall configuration for Copilot cloud agent is inherited and can be managed at organization level.
Default state
Conditional
Availability
As documented on 2026-09-09, automations are available in private and internal repositories only, and not in public repositories, in repositories owned by managed user accounts, where a repository administrator has disabled them, or where Copilot cloud agent is not enabled. The organization must allow both Copilot cloud agent and automations in the repository, both of which are enabled by default. Any user with write access to a repository can create automations in that repository. Automations can be created and managed from the Agents tab in a repository on GitHub in the Automations pane, from the Automations tab in the GitHub Copilot app, and from a user-level Automations view across repositories. The rationale, confidence and approvals behaviour for issue changes is documented as in public preview and subject to change.
Licensing
GitHub documents automations as available with the GitHub Copilot Pro, GitHub Copilot Pro+, GitHub Copilot Max, GitHub Copilot Business and GitHub Copilot Enterprise plans. Each automation run starts a Copilot cloud agent session that uses GitHub Actions minutes and GitHub AI Credits, billed to the user who created the automation. No pricing is recorded here.
External model or provider
Not publicly established for this capability. GitHub documents that the user optionally selects the model Copilot uses when creating an automation, referring to the guidance on changing the AI model for Copilot cloud agent, but the current automations documentation does not identify a specific model or third-party provider for this capability. Represented conservatively as a variable, user-selectable model configuration with provider identity not established.
Limitations and uncertainty
Execution is automatic only within boundaries a user configures in advance: the prompt, triggers, optional model, selected tools and the single repository the automation is scoped to. The selected tools are the main control over what actions Copilot can take, so an automation does not automatically receive every tool. GitHub states that automations run Copilot without a person initiating each task and therefore carry some additional risks, and documents mitigations: attribution of pull requests and pushed code to the automation creator (who cannot approve those pull requests), least-privilege tool selection, ignoring events triggered by users without write access by default to reduce prompt-injection risk, and GitHub Actions workflows not running on a pull request until a user with write access approves them. For issue changes only, an automation explains each change and rates confidence as high, medium or low; changes below the repository automation level threshold are held as suggestions for human accept or decline, and a prompt can explicitly request suggestion rather than application. GitHub states these approvals are not a security control and do not enforce a server-side boundary. This confidence and approval mechanism applies only to changes to issue labels, fields, issue type, closing issues and assignees, and not to other actions such as opening pull requests or pushing code. Automations are private to their creator, while the sessions they start are visible to people with repository access. Automations are not versioned in Git. The runtime permission model beyond creator write access, attribution and selected tools is not fully established, and whether inherited firewall and configuration permit external reach in a given repository depends on configuration. This record covers automations that run through Copilot cloud agent; GitHub Agentic Workflows and other agent automation mechanisms are treated separately.

Evidence