test(sandbox): exercise a sandbox session end to end

[ItamarZand88]

Summary

Exercises a sandbox end to end: the test app declares one, and the suite drives a real session through it rather than asserting on rendered templates.

What the test does when it runs:

1. Deploys the comprehensive test app, which now declares a sandbox alongside its other resources.
2. Creates a session through the binding and runs a command in it.
3. Writes a file, reads it back, and checks the content — which is the only thing that proves the agent, the capability and the transport all line up.
4. Terminates the session and confirms it is gone.

What I did

The assertions are on observed behaviour, not on artifacts. A test that checks a template contains a string would pass against a sandbox that never starts, so this one runs code inside the session and reads back what it wrote.

Files touched

• crates/alien-test/ — the sandbox case in the deployment suite.
• tests/e2e/test-apps/comprehensive-typescript/ — the declaration and the handler the test drives.

How I tested

The suite itself is the test: it creates a session, runs a command, round-trips a file and
terminates, asserting on what came back rather than on rendered artifacts.

I have not run it against a live deployment in this change — it needs cloud credentials and a
free slot. It should be run before merge, and the thing to watch is teardown: an accepted delete
is not a completed one, so confirm the session is actually gone rather than trusting the call.

[greptile-apps]

Greptile Summary

The PR adds end-to-end sandbox coverage to the comprehensive Rust and TypeScript test applications.

• Declares and links a Local-only sandbox with execution permissions.
• Exercises command execution and file round-tripping through a real session.
• Terminates the session and verifies that cleanup completed.
• Registers the sandbox binding in the shared E2E runner.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains; the previously reported false-success cleanup path now runs termination after the exercise and propagates cleanup failure.

Important Files Changed

Filename	Overview
tests/e2e/test-apps/comprehensive-rust/src/handlers/sandbox.rs	Adds a complete Rust sandbox exercise and now propagates cleanup or termination-confirmation failures.
tests/e2e/test-apps/comprehensive-typescript/src/handlers/sandbox.ts	Adds the equivalent TypeScript sandbox exercise while ensuring cleanup runs even when functional checks fail.
crates/alien-test/tests/common/bindings.rs	Adds the runner-side sandbox endpoint check and validates the successful binding response.
crates/alien-test/src/e2e.rs	Registers sandbox as a supported binding for the Local deployment model.

Sequence Diagram

sequenceDiagram
    participant Runner as E2E runner
    participant App as Comprehensive test app
    participant Sandbox as Sandbox binding
    Runner->>App: POST /sandbox-test/alien-sandbox
    App->>Sandbox: create session
    App->>Sandbox: run command
    Sandbox-->>App: stdout and exit frame
    App->>Sandbox: write file
    App->>Sandbox: read file
    Sandbox-->>App: matching bytes
    App->>Sandbox: terminate session
    App->>Sandbox: get session
    Sandbox-->>App: absent or terminated
    App-->>Runner: success

Loading

_{Reviews (6): Last reviewed commit: "test(sandbox): exercise a sandbox sessio..." | Re-trigger Greptile}

Dismissed because a newer commit was pushed; Greptile will re-review the current head.