Skip to content

Tools

Siri AI opens custom app actions and multi-step skills in developer preview

Edited and reviewed by AI Ping Editorial Team

3-line summary

  1. Apple’s 27.3 developer betas preview custom actions and content from third-party apps.
  2. App skills guide multi-step requests using functionality and data developers expose.
  3. Testing is internal; Apple plans to open App Store submissions in the spring.
PracticalDevelopers
A phone with an abstract voice waveform beside a notebook and test cable
AI-generated conceptual image: a phone workspace for testing voice integrations · AI Ping

Apple opened a developer preview on November 3 that lets third-party apps expose custom actions and content to Siri AI. The iOS, iPadOS, macOS and visionOS 27.3 betas also introduce app skills to guide requests involving several actions.

The change expands integration beyond predefined categories of actions and content. Developers can build and test their own integrations now; Apple says customer availability will follow in a few months, with App Store submissions opening in the spring.

One example in Apple’s documentation starts with shoes saved in a shopping app. Siri could retrieve that information, identify the user’s partner through Contacts and send it through a specified messaging app. This is Apple’s proposed workflow, dependent on the actions participating apps expose.

A phone with an abstract voice waveform between sneakers and a paper envelope
AI-generated conceptual image: objects representing a request to find and share saved shoe information · AI Ping

What a person means also depends on the screen they are viewing. Apple describes associating visible content with app entities so Siri can understand a reference such as “this photo.” The contextual-cues framework covers files, documents and other visible content. Those associations supply information that would otherwise remain private to the app.

Skills guide a workflow; apps define access

App skills supply the domain knowledge and guidance needed to combine individual actions. Apple’s shopping example covers finding products, filling a cart, choosing payment and delivery methods, reviewing the cart and confirming a purchase. Guidance can explain which action to prefer and what needs clarification.

The skills are Markdown files validated and compiled by Xcode, then indexed on the device. Siri checks their names and descriptions before loading relevant instructions. Apple says skills cannot enforce deterministic behavior: the assistant does not simply execute a listed sequence. They provide context for choosing and coordinating tools.

Developers explicitly expose actions through App Intents and describe the app’s data. Passing information between apps also requires a transferable representation the receiving app can handle.

Authentication requires deliberate configuration. Existing policies can allow execution even on a locked device, require authentication, or require the executing device to be unlocked. The selected policy therefore matters alongside which action is exposed. Apple’s implementation guide says app entities require no authentication by default before their information is presented. Developers can require it for an entity or individual properties, and mark properties internal so Siri and Shortcuts cannot read or surface them. Actions returning sensitive results need intent-level authentication.

Testing is restricted to development-signed builds on supported devices and internal TestFlight testers. The custom integrations are unavailable to external TestFlight testers and App Store users during this preview. The APIs and Siri behavior may change.

Diagram of current developer and internal testing access and planned App Store submissions
Illustration: internal preview access and planned spring App Store submissions · AI Ping

The accompanying Siri Evaluations framework scripts conversations and compares the actions called, parameter values, information found and response text against expectations. Checks can also cover which skills load and which entities appear onscreen. They are designed to catch changes as developers revise their integrations.

Xcode reports what Siri says and does at each turn. Apple advises testing indirect, incomplete and unexpected wording across repeated runs because responses can vary. The checks themselves use deterministic comparisons; that property does not make the assistant’s workflow deterministic.

Access remains subject to device, language, regional and usage conditions. Apple’s existing Intelligence terms say reaching a limit can restrict a feature until its allowance resets. The company plans to provide further terms and usage-policy details for the new third-party Siri experiences.

Sources