apple-intelligence

What the source does

• FilesystemNone observedNo reads or writes in the source
• NetworkNone observedNo outbound calls in the source
• ShellNone observedNo commands run in the source
• DependenciesPinnedNothing floats
• SecretsNone foundNo keys in source

Five audit layers

Gates applied: no behavioural pass.

- Apple-trained guardrails apply to **both input and output**: "Your instructions, prompts, and tool calls are all considered inputs... harmful model outputs are blocked, even if the prompts were craf

- Apple-trained guardrails apply to **both input and output**: "Your instructions, prompts, and tool calls are all considered inputs... harmful model outputs are blocked, even if the prompts were craf

Audit 0.4.1 · source sha 7fd70e317e5c · 26 Oct 2026Report an issue or request a re-scan

• Publisherrshankras GitHub userOwner of the repository this skill is read from
• Source folderrshankras/claude-code-apple-skills/skills/apple-intelligenceThe exact folder we audited
• LicenceMITAs the repository states it
• Commit pinned for this verdict7fd70e317e5cA new audit runs when the source changes
• Repository statusNot archivedLast push 93 days ago
• Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

Its description: Apple Intelligence skills for on-device AI features including Foundation Models, Visual Intelligence, App Intents, and intelligent assistants. Use when implementing AI-powered features.

---
name: apple-intelligence
description: Apple Intelligence skills for on-device AI features including Foundation Models, Visual Intelligence, App Intents, and intelligent assistants. Use when implementing AI-powered features.
allowed-tools: [Read, Write, Edit, Glob, Grep, Bash, AskUserQuestion]
last_verified: 2026-07-16
review_by: 2027-06-22
os_version: iOS 27 / macOS 27
---

# Apple Intelligence Skills

Skills for implementing Apple Intelligence features including on-device LLMs, visual recognition, App Intents integration, and intelligent assistants.

## When This Skill Activates

Use this skill when the user:
- Wants to add AI/LLM features to their app
- Needs on-device text generation or understanding
- Asks about Foundation Models or Apple Intelligence
- Wants to implement structured AI output
- Needs prompt engineering guidance
- Wants camera-based visual intelligence features
- Needs Siri, Shortcuts, or Spotlight integration via App Intents
- Wants to expose app actions or content to the system

## Available Skills

### foundation-models/
On-device LLM integration with prompt engineering best practices.
- Model availability checking
- Session management
- @Generable structured output (property-order + schema-injection rules)
- Tool calling patterns
- Snapshot streaming
- Prompt engineering techniques
- `safety-and-guardrails.md` — model limits, the instructions-over-prompts hierarchy, guardrail handling, the four-layer safety stack, evals
- `models-and-agents.md` — Private Cloud Compute, LanguageModel protocol (third-party backends), vision input, DynamicProfile agents, tool-calling modes, KV-cache rules

### visual-intelligence/
Integrate with iOS Visual Intelligence for camera-based search.
- IntentValueQuery implementation
- SemanticContentDescriptor handling
- AppEntity for searchable content
- Display representations
- Deep linking from results

### app-intents/
App Intents for Siri, Shortcuts, Spotlight, and Apple Intelligence.
- AppIntent protocol, parameters, perform()
- AppEntity and entity queries
- App Shortcuts with voice phrases
- IndexedEntity and Spotlight indexing
- Intent modes (background, foreground)
- Interactive snippets with SnippetIntent
- Visual intelligence integration
- Onscreen entities for Siri/ChatGPT
- Multiple choice API
- Swift package support

## Key Principles

### 1. Privacy First
- On-device processing by default; Private Cloud Compute extends capacity without storing prompts (no account/API keys — WWDC26 241)
- When routing to less-private backends, redact context first (see foundation-models/models-and-agents.md)

### 2. Graceful Degradation
- Always check model availability
- Provide fallback UI for unsupported devices
- Handle errors gracefully (incl. guardrail violations — silent for proactive features, explained for user-initiated)

### 3. Efficient Prompting
- Keep prompts focused and specific
- Use structured output when possible
- Respect context window limits — query `model.contextSize`, never hardcode (4,096 first-gen on-device; 8,192 iOS 27 on-device; 32,000 Private Cloud Compute)

## Reference Documentation

- [Foundation Models](https://developer.apple.com/documentation/foundationmodels)
- [Visual Intelligence](https://developer.apple.com/documentation/visualintelligence)
- [App Intents](https://developer.apple.com/documentation/appintents)
- Local captured docs (optional): if `~/Downloads/docs/` contains `FoundationModels-Using-on-device-LLM-in-your-app.md`, `Implementing-Visual-Intelligence-in-iOS.md`, or `AppIntents-Updates.md`, read them for extra detail; skip silently if absent.

Excerpt, first 3,607 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

git clone https://github.com/rshankras/claude-code-apple-skills
git -C claude-code-apple-skills checkout 7fd70e317e5c

Works in

1. 2026-10-26
   ✓SAFE B 89 / 100

   first audit · source 7fd70e317e5c

Questions about apple-intelligence

• github.com/rshankras/claude-code-apple-skills/tree/main/skills/a — the audited source at commit 7fd70e317e5c and every finding aboveObserved · 26 Oct 2026

Spotted something wrong? Report a correction · Methodology