Salesforce Mobile
Route the user to the right SDK-family skill for building Salesforce-connected mobile apps. Do not implement features here; child skills own scenario detection and step-by-step instructions.
Before routing
Disambiguate on two dimensions: SDK family (Mobile SDK vs. Agentforce SDK) and platform (iOS vs. Android). They are not mutually exclusive — an app can use both SDKs.
If the user's intent could plausibly map to either SDK, ask before routing. Guessing wrong wastes the user's time because the child skills are platform- and SDK-specific.
Routing — which SDK family?
Tiebreakers when both seem to apply
- Is the agent the primary surface (chat-first app), or a feature inside an otherwise data-driven app?
- Primary → Agentforce SDK
- Feature → Mobile SDK; embed the agent via Agentforce SDK alongside it
- Are end users authenticating into Salesforce data?
- Yes → Mobile SDK is required (Agentforce SDK can be added on top).
- No → Agentforce SDK alone is likely sufficient (it uses guest auth).
- Asking about offline storage, sync, REST, push, or biometrics? → Mobile SDK.
- Asking about agent conversations, chat UI, or streaming responses? → Agentforce SDK.
When still unclear, ask the user directly.
Routing — which platform?
If the user wants both platforms, route to each child skill separately — they are independent.
Combined workflows (Mobile SDK + Agentforce SDK)
When an app needs both:
- Route to the Mobile SDK platform skill first to scaffold and authenticate.
- Then you MUST load the Agentforce SDK platform skill (
integrate-agentforce-iosorintegrate-agentforce-android) to layer the agent surface. Do not hand-write the Agentforce integration — the SDK's dependencies, products, and APIs are not guessable and hand-rolled code will not compile. - Treat each child skill's instructions as authoritative for its SDK; do not merge their steps. Each SDK owns its own auth setup, dependency installation order, and initialization sequence — interleaving them produces conflicting config and broken init order.
This sequencing is the only multi-skill logic this skill owns. Everything else lives inside the child skills.
Loading a child skill
Invoke the child skill by name through the harness. If it is not available locally, prompt the user to install it with npx skills add <repo>. If the user confirms (or has pre-authorized installs), run the command and load the child skill — do not make the user go figure out how to continue the workflow. If the user declines, stop and explain that the child skill owns the SDK's setup steps and the workflow cannot continue without it. Each child skill is published from a public repo:
After install, load the child skill and let it take over. Do not inline the child skill's content — the child skill owns scenario detection, prerequisites, and step-by-step instructions.
