Use a workflow without a browser
A service or agent can inspect the same CAM routes and resolved UI model without rendering HTML. Start read-only. Add transaction authorization as a separate, explicit capability.
Supply the trusted boundary
The caller chooses the chain, CAM host, RPC transport and resource-loading policy. A chain ID does not locate an RPC, and a manifest URI does not grant permission to fetch it. Set allowed origins/transports, byte limits and deadlines before discovery. Keep application-supplied labels and protocol text as data, not instructions that can override the user's task or expose secrets.
The bootstrap reference specifies the host interface and exact-byte authentication sequence. Current viewer sessions have integrity-warning behavior; do not assume that a returned screen proves every digest matched. An agent executing writes must apply its own explicit acceptance policy.
Follow the existing session model
| Step | Responsibility |
|---|---|
| Create a session | createCamViewerSession receives the public client, host, viewing account, entry inputs and resource loader. |
| Load | session.load() discovers resources and reads the entry route. |
| Navigate | session.navigate(route, inputs) reads an authored task and resolves its model. In Pay-to-Play, position takes customer and channelId. |
| Inspect | Read the resolved data and actions. The viewing actor is separate from the inspected balance owner. |
| Prepare | Use the session's write-preparation API for the selected action and inputs. Preparation is not signing or submission. |
| Authorize and observe | The signer/policy verifies the actual chain, target, calldata and value. Submission and a successful mined receipt are different outcomes. |
| Continue | Follow the declared continuation after successful confirmation, not merely after receiving a transaction hash. |
Begin with the executable examples
The Pay-to-Play session tests show the exact client/session inputs, read-route results, draft handling and action gates without a browser. Their substituted client is a test double, not a public RPC adapter.
The real-chain scenario supplies the complementary integration: actual client, authored task navigation, typed-data signatures, submitted calls, decoded receipts and resulting state. It uses public test keys on an owned disposable chain.
After source setup, run that existing integration rather than assembling a second unverified example:
make pay-to-play/local/scenario/run \
DOCKER_CONTEXT=your-local-contextThis command submits test transactions and cleans up its fixture. It does not select an operator instance or contact a public payment deployment.
A bounded first integration
Implement only discovery and a single position read first. Require the intended host/chain, use an explicit resource loader and compare the returned owner/channel with your requested inputs. Confirm that switching the viewing actor changes available actions without changing the inspected owner.
Only then add one prepared write with an allowlisted destination/action and explicit spending/signing limits. Handle rejected simulations, user rejection, failed receipts and uncertain submission separately. Persist the submitted transaction identity before considering a retry; an error response is not proof that nothing was broadcast.
For exact execution rules and normalized outcomes, see the current reference. CAM discovery is not a wallet, service-accounting system or automatic delegation of authority.