Connect your application

Choose the correct Dino boundary for input, in-call work, and terminal results.

Data flow

Your contact data
   ├──▶ Placeholders ──▶ Model-visible conversation values
   └──▶ Plugin context ──▶ One addressed integration only

Model tool call ──▶ Agent-owned integration ──▶ External system

Terminal call ──▶ Process record or signed Drip callback ──▶ Your application

Choose the boundary

Need	Use
Mention a value during the conversation	Declared placeholder
Give trusted private data to one integration	Integration context
Validate or update an external system during the call	In-call integration tool
Record a structured result in Dino	Post-call capture tool
Receive each Drip outcome in your application	Signed callback
Recover a failed Drip delivery	Failed-callback queue

Prompt your AI agent

Recommended

Use the `dino-ping` and `dino-pong` skills to design my application boundary.
My application supplies: <data>.
The model may see: <allowed data>.
The external side effects are: <effects>.
The terminal results are: <results>.
Classify each field as placeholder, private integration context, tool argument, or callback result.
Highlight any field that crosses the wrong trust boundary.
Do not change an agent or process.

Use the Dino CLI directly

dino agent get --agent-id <agent-id>
dino agent validate --agent-dir <absolute-agent-dir>
dino process get --process-id <process-id>
dino process get-call --process-id <process-id> --call-id <call-id>

Gotchas

• Do not put private integration context into model-visible placeholders.
• Do not ask the model to invent trusted record or context fields.
• Keep static integration configuration non-secret.
• Evolve the agent to change integrations. Processes do not attach plugins.
• Keep normal callback handling and failed-delivery reconciliation idempotent.

Next

Use In-call tools for live external work or Get call results for terminal data.