What does this skill do, and when should you use it?
This skill helps developers analyze Android apps to identify key user workflows that can be exposed as AppFunctions, and generates the corresponding Kotlin code. AppFunctions are part of the Android platform API (available from Android 16) that allow system agents to invoke app features directly without opening the UI. The skill covers the full flow from discovery, implementation, KDoc refinement, to ADB-based testing. It requires apps to target SDK 36+ and compile with SDK 37+, and recommends using the Jetpack library for backward compatibility.
Analyzes Android codebases to identify and recommend candidate AppFunctions (e.g., creating notes, playing media, sending messages). Generates Kotlin implementations for AppFunctions and configures system-wide and build dependencies. Refines KDoc documentation to ensure AI agents and MCP can correctly understand and use the functions. Provides ADB commands to list, invoke, and verify registered app functions.
- A developer wants to let an AI assistant trigger app actions like creating a note or playing media without opening the app.
- An Android app needs to expose its features as AppFunctions for system agents to discover and execute on-device.
- A team wants to improve the KDoc of their function APIs so that LLMs and MCP clients can use them accurately.
- A developer needs to debug or verify that their AppFunctions are properly registered using ADB commands.
- A developer migrating from an older AppFunctions version (≤ alpha09) to the new service entry point architecture.
How do you install this skill?
- The skill may depend on overseas services (e.g., Google Maven, Androidx) that may be unreachable in mainland China networks; environment fit should be carefully evaluated.
- No version history or changelog provided, and maintenance responsibility is unclear, posing long-term update risks.
- Static review cannot verify actual execution; reliability and effectiveness scores are inferred from documentation, so testing is needed before practical use.
- Shell / CLI
- Local filesystem
KotlinKSPADB
Use Android CLI to install this skill into the current directory: android skills add --skill=appfunctions --project=.. To install all skills, run android skills add --all. If no agent directories are specified, skills are installed to ~/.gemini/antigravity/skills.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/android/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/device-ai/appfunctions" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- discover AppFunctions in my app
- implement AppFunctions and configure Hilt
After installation, ask your AI assistant to perform tasks like 'discover AppFunctions in my app' or 'implement AppFunctions and configure Hilt'. The skill will guide you through the steps. Ensure your app meets prerequisites: targetSdk 36+, compileSdk 37+ and uses the Jetpack library for AppFunctions.
What are this skill's strengths and limitations?
- Follows official Android platform standards and Google best practices
- Covers the complete development flow from discovery to testing with clear steps
- KDoc refinement improves AI and MCP understanding
- Provides ADB commands for local verification
- Requires high SDK levels (targetSdk 36+) which may be costly for legacy apps
- Only works on Android 16+; the Jetpack library is required for backward compatibility
- No bundled test cases or example code; verification is left to the user
- May not cover all app types or complex scenarios
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| AppFunctions Skill this page | 61 · Recommended | ★ 7.7k | 16d ago | Apache-2.0 |
| Agent Reach | 62 · Recommended | ★ 94k | 4d ago | MIT |
| skill-optimizer: Eval Workbench for Agent Skills | 61 · Recommended | ★ 81 | 5mo ago | MIT |
| CameraX Development Skill | 60 · Recommended | ★ 7.7k | 16d ago | Apache-2.0 |
| Expo Native UI Builder | 59 · Recommended | ★ 2.7k | 6d ago | MIT |
How did FollowSkills review this skill?
The skill explicitly states security constraints to avoid exposing sensitive data and to require user confirmation for destructive actions, and follows least privilege principles; no malicious or covert behavior detected, red lines not triggered. Deductions for incomplete permission management and confirmation process details.
Instructions and reference docs are internally consistent and provide detailed steps and examples, but no test suites or reproducible execution evidence; error handling is limited to a few troubleshooting tips and does not cover all edge cases. Deductions for lack of tests and comprehensive error handling.
The skill has clear usage scenarios and trigger conditions, and a defined target audience, but capability boundaries and non-fit scenarios are not explicit, and no adaptation for mainland China network environments; may rely on overseas services. Deductions for unclear boundaries and environment fit.
Documentation is well-layered, includes prerequisites, steps, examples, and troubleshooting, and provides LICENSE and author info, but lacks explicit version history, changelog, and maintenance responsibility statements. Deductions for incomplete version governance.
The skill clearly describes the core task and expected outputs, but direct usability of outputs is unverified, and comparative benefit over alternatives is unclear. Deductions for lack of actual validation.
Key claims are supported by documentation, but only a single official source; no third-party corroboration or executable tests/CI evidence. Deductions for limited evidence types.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →