A2A Protocol¶
Video Intelligence Agent is built on the Agent-to-Agent (A2A) Protocol v1.0 — an open standard for enabling seamless, interoperable communication between AI agents and the clients or platforms that talk to them.
What is the A2A Protocol?¶
The A2A Protocol defines a standardized way for AI agents to:
- Advertise their identity and capabilities (via an Agent Card)
- Accept requests from any compatible client or platform
- Manage tasks through a well-defined lifecycle (Submitted → Working → Completed)
- Return results as structured artifacts
- Stream updates in real time as work progresses
Because Video Intelligence Agent is A2A-compliant, it can be invoked by any client, orchestrator, or platform that understands the A2A standard — no custom integration code required.
Agent Card¶
Every A2A agent publishes an Agent Card — a machine-readable document that describes what the agent can do. Video Intelligence Agent's Agent Card is automatically served and contains:
| Field | Value |
|---|---|
| Name | Video Intelligence Agent |
| Version | 1.0.0 |
| Protocol Version | 1.0 |
| Streaming | Enabled |
| Provider | LTIMindtree |
Accepted Input¶
| Format | Description |
|---|---|
| MP4 video | Standard video format |
| WebM video | Open web video format |
| QuickTime video | Apple video format |
| Plain text | Optional context alongside the video |
Produced Output¶
| Format | Description |
|---|---|
Gherkin .feature files |
One per identified business domain |
summary.json |
Metadata: total features, scenarios, flows identified |
Skill¶
Video Intelligence Agent exposes a single skill:
BDD Test Case Generation from Video — Accepts a video demonstrating application workflows and generates comprehensive Gherkin feature files covering all visible user flows.
Example prompts recognized by this skill:
- “Generate BDD test cases from this application demo video”
- “Analyze this screen recording and create Gherkin feature files”
- “Watch this video and write comprehensive test scenarios”
Task Lifecycle¶
Every request to Video Intelligence Agent is tracked as a task with a defined lifecycle:
stateDiagram-v2
[*] --> Submitted : Request received
Submitted --> Working : Analysis begins
Working --> Working : Incremental status updates
Working --> Completed : All artifacts delivered
Working --> Failed : An error occurred
Working --> Canceled : Client canceled the request
Completed --> [*]
Failed --> [*]
Canceled --> [*]What Each State Means¶
| State | What is happening |
|---|---|
| Submitted | The request has been received and the task has been created |
| Working — Analyzing video | The video has been forwarded to Gemini AI for multimodal analysis |
| Working — Generating feature files | Gemini has responded; Video Intelligence Agent is assembling the feature files |
| Completed | All feature files and metadata have been delivered |
| Failed | An unrecoverable error occurred; a descriptive message is returned |
| Canceled | The client requested cancellation before completion |
Communication Modes¶
Video Intelligence Agent supports two modes of communication:
| Mode | Description | Best For |
|---|---|---|
| Send (non-streaming) | Single request — single response after full completion | Programmatic clients |
| Stream (SSE) | Real-time event stream — each status update and artifact arrives as it is ready | Interactive clients, dashboards |
Security¶
Video Intelligence Agent uses standard Google identity-based authentication:
| Scheme | Description |
|---|---|
| Google OpenID Connect | Authenticate via your Google account. Obtain a token using gcloud auth print-identity-token. Pass as a Bearer token. |
| HTTP Bearer (JWT) | Google-issued JWT for Cloud Run IAM or Cloud Marketplace authentication. |
Health Checks¶
Two health-check endpoints are available outside the A2A protocol for infrastructure monitoring:
| Endpoint | Response |
|---|---|
GET /health |
{"status": "ok"} |
GET /healthz |
{"status": "ok"} |