FoneClaw Tools, Plugins, Skills, Workflows, and Shortcuts Explained

Tools vs Plugins vs Skills vs Workflows in One View

The fastest way to understand FoneClaw tools plugins skills workflows is to give each layer one owner verb. A Tool acts. A Plugin adds. A Skill guides. A Workflow sequences. A Shortcut starts. We use these layers together because an Android phone agent needs more than a single abstraction. It needs atomic execution, installable capability, reusable know-how, repeatable procedure, and fast entry points.

These layers compose rather than compete. A shortcut can start a workflow. A workflow can use a skill for guidance. The agent can select built-in tools or plugin-contributed tools. Runtime controls still decide what can execute and what needs approval. The public FoneClaw Features page gives the user-facing capability overview, while this guide owns the vocabulary and decision model.

Atomic Tools and Installable Plugins

Tools are the execution atoms in FoneClaw. A tool has a name, purpose, inputs, output, policy metadata, and runtime controls. We design tools this way because Android phone actions need clear contracts. Opening an app, checking Bluetooth state, adjusting supported volume, reading visible screen context, preparing a task, or applying a supported setting should have a defined action boundary. That boundary lets the agent plan, lets the user review important steps, and lets the system record what happened.

FoneClaw exposes 100+ built-in tools across governed capability areas. We use that stable phrase because the catalog can evolve while the product principle stays the same: each action should be explicit, scoped, and governed. Tool management includes visibility and approval controls, so execution is not just a model decision. Android permissions, FoneClaw approval settings, and the task state all matter at runtime.

Plugins are different. A Plugin adds installable capability. It can deliver its own Android package, signer identity, permissions, metadata, and contributed tools. The host verifies package metadata and expected trust signals before registering plugin tools. Once installed, plugin tools still run through contracts, permission checks, and approval behavior. A Plugin expands what FoneClaw can do; it does not erase runtime authority checks.

File operations make that concrete. AI File Manager Agent Android Plugin: Safe File Operations With FoneClaw shows how a plugin-backed file capability needs path scope, previews, approvals, and observed outcomes. Media tasks offer another example: Free FoneClaw and Free Local YouTube Downloader Plugin for Android demonstrates how an installable plugin can provide a service-specific capability while remaining separate from built-in tools. The common pattern is capability plus contract.

Guidance, Repeatable Procedures, and Fast Entry Points

Skills, Workflows, and Shortcuts shape behavior above the raw action layer. A Skill is reusable guidance. It can include metadata and instructions that help the agent recognize and handle a class of tasks. In our product vocabulary, a Skill guides task handling; it does not directly grant Android authority. When the moment comes to act, the runtime still uses tools, permissions, approvals, and result checks.

A Workflow saves an ordered procedure. It is useful when the same type of task happens repeatedly: prepare for a meeting, open the right app, set a supported device state, check a visible screen, create a reminder, or perform a sequence that has known checkpoints. A Workflow can reduce repeated prompting because the user has already shaped the procedure. It still remains subject to tool contracts and approval behavior. A workflow is repeatable; it is not automatically unattended.

A Shortcut is a fast entry point. It starts a known path without making the user restate the whole request. That might be a command for meeting prep, a quick device state, a frequent app-opening routine, or a saved task start. Shortcuts matter because phone-agent interaction often happens in motion. The user wants a compact way to start something familiar, then see the relevant state or approval when needed.

The same goal can appear at all three layers. Suppose the task is "prepare for my weekly call." A Skill can guide how meeting preparation should be interpreted. A Workflow can save the ordered steps: set volume, apply Do Not Disturb if approved, open the calendar, and prepare a note. A Shortcut can start that workflow quickly. The underlying tools still perform the actual Android actions. For broader examples of saved multi-step routines, Automate Android Tasks With One Voice Command is the natural next guide.

How FoneClaw Capability Layers Compose

Composition starts with a request and context. The user asks for an outcome, attaches current-screen context when helpful, or starts from a shortcut. The configured model interprets the request, then the agent checks whether a Skill provides useful guidance, whether a Workflow already captures the procedure, which Tool can act, and whether a Plugin contributes a relevant capability.

Authority stays with the layer that owns the action. A Skill can recommend how to handle a request, but it does not approve a message or grant file access. A Workflow can sequence a task, but each tool step still has its own policy. A Plugin can add tools, but Android permissions and FoneClaw approvals still apply at execution time. A Shortcut can launch a known route, but it does not turn a sensitive operation into a silent action.

A realistic composition might look like this: the user taps a shortcut for meeting prep. The workflow opens, a skill guides the preparation style, the agent checks Bluetooth and volume with built-in tools, a supported Do Not Disturb action requests approval, and the result is verified. If a plugin-backed file task is part of the workflow, the plugin tool contributes that capability under its own contract. The user sees the meaningful state changes and recovery options.

This is the same broader shift happening across machine-callable app ecosystems. App Intents and Machine-Callable Apps for AI Agents explains how external apps can expose structured capabilities to agents. FoneClaw's internal layer model follows the same principle: use explicit capability contracts, compose them around intent, and keep authority clear.

Permissions, Approvals, and Data Ownership Across Layers

Permissions, approvals, and data ownership do not collapse into one switch. Android permissions decide whether the phone allows access to a surface such as files, contacts, screen content, microphone, notifications, Bluetooth, or settings. FoneClaw approvals decide whether a specific action should proceed in the current task. Plugin contracts describe what a contributed capability needs and what it returns. Skills and Workflows store guidance and procedure, not open-ended authority.

Installed capability still needs runtime checks. A plugin may request Android permissions as part of its package and expose tools after verification. That installation step tells FoneClaw what capability exists. The actual task still needs the right input, the right permission state, and the right approval when the effect matters. Sending, deleting, changing device state, downloading, writing, or sharing should remain tied to the visible task.

Data ownership also varies by layer. A Skill may contain reusable instructions. A Workflow may store ordered steps and defaults. A Plugin may process data needed for its specific capability and may have its own service or storage behavior depending on its contract. FoneClaw's job is to keep scope visible: what context is used, what tool is called, what permission is required, what result returns, and what the user can remove or stop.

Security analysis deserves its own depth, so AI Agent Skill Security Needs Phone Permission Checks covers the permission side more fully. The product rule here is practical: choose the smallest layer that owns the change, then let permissions, approvals, and result checks do their work.

Three End-to-End FoneClaw Examples

Start with a reversible setting task. The user asks FoneClaw to prepare a meeting mode. A Shortcut can start the familiar path. A Workflow can sequence volume, Do Not Disturb, and screenshot or note setup. Built-in Tools inspect and adjust supported device state. Approval appears when the action changes how the phone handles interruptions. Stopping and permission recovery stay available through the current task. The current released FoneClaw baseline supports this behavior with floating assistant access, current-screen attachment, task continuity, approvals, stopping, permission recovery, and quick actions.

The second example is a repeatable work routine. A user wants to prepare for a daily standup: open calendar context, create a short memo, check notifications, and launch the relevant app. A Skill can guide what "standup prep" usually means. A Workflow saves the ordered steps. Tools perform app opening, memo, task, notification, or settings actions where supported. A Shortcut starts the routine from a known entry point. The result is not one giant tool; it is layered composition.

The third example uses Android agent plugins. Suppose the user needs a file task before sending a status update. A File Manager Plugin can contribute file operations under its own package and permission contract. The Workflow can include a file preview step, then the user can approve the relevant action. AI File Manager Agent Android Plugin: Safe File Operations With FoneClaw gives the deeper case study because file tasks need path scope, preview/apply separation, and clear outcomes.

These examples show the model we build toward: Tools execute, Plugins add capability, Skills guide, Workflows sequence, and Shortcuts start. The public FoneClaw Features page summarizes the current user-facing capability surface, including 100+ built-in tools, while this guide explains how to reason about the layers behind it.

Choose the Right FoneClaw Extension Layer

Use this decision tree when deciding where a capability belongs. If you need one specific phone action, use or define a Tool. If you need a new installable service or device capability, use a Plugin. If you need reusable know-how for a class of tasks, use a Skill. If you need a repeatable ordered procedure, use a Workflow. If you need a faster way to start a known route, use a Shortcut.

Common mistakes come from assigning ownership to the wrong layer. Turning every instruction into a tool makes execution vague. Treating a Skill as action authority skips runtime controls. Using a Workflow for a one-time task adds needless structure. Installing a Plugin for a capability already covered by built-in tools increases surface area. Creating a Shortcut before the procedure is stable can make errors faster too.

The best starter path is reversible. Create or test a simple workflow for a supported setting or reminder. Add a shortcut only after the steps feel repeatable. Add a skill when the task needs reusable guidance. Add a plugin when the Android capability truly comes from an installable package. Keep approvals and permission recovery visible while you test.

For the broader ecosystem context, return to App Intents and Machine-Callable Apps for AI Agents. FoneClaw's internal capability layers and external machine-callable app capabilities share one important idea: agents become more useful when capabilities are explicit, scoped, and composed around the user's intent.