Every test script runs with three globals available: test(), expect()/assert(), and res. This page is the full reference for all three.

test(name, fn)

Registers one named check.

test('a descriptive name', () => {
  // assertions go here
})

• fn can be synchronous or return a Promise — an async check is supported:

  test('async check', async () => {
    expect(res.getBody().status).to.equal('ok')
  })
  

• If fn throws (or its returned promise rejects), the test is recorded as failed, with the thrown error’s message. If Chai’s assertion carries .actual/.expected (e.g. to.equal, to.eql), both are captured for the result view’s side-by-side display.
• If fn returns without throwing (or its promise resolves), the test is recorded as passed.
• A script can call test() any number of times; each call is an independent row in the results.

expect() / assert()

Both are Chai’s assertion APIs, available as-is:

expect(res.getText()).to.include('success')
expect(res.raw.content).to.have.lengthOf(2)
assert.isTrue(res.isError === false)

Anything Chai’s expect/assert support is available — .to.equal, .to.eql, .to.include, .to.be.a(...), .to.throw, chained matchers, and so on.

console.log / .info / .warn / .error

All four are captured (not printed anywhere real) and shown in a collapsible Console section in the result view — useful for inspecting intermediate values while writing a script.

console.log('body was', res.getBody())

res — the call result

res’s shape depends on which kind of capability the script is attached to (tool, prompt, or resource) — set automatically based on where the Tests tab lives.

Shared across all three kinds

Member	Description
res.getText()	All text content joined with \n (empty string if none)
res.body / res.getBody()	The parsed “payload” — see per-kind notes below
res.responseTime / res.getResponseTime()	Call duration in ms, or null if not measured
res.getSize()	{ body, total } — real UTF-8 byte counts
res.raw / res.getRaw()	The whole underlying result object, untouched

Tool results

Member	Description
res.isError / res.getIsError()	true if the tool call reported an error
res.content / res.getContent()	The raw content array (or [])
res.getText()	All text-type content blocks joined with \n
res.body / res.getBody()	structuredContent if present; otherwise getText()’s output, JSON-parsed if it parses, else the raw string, else null

Prompt results

Member	Description
res.messages / res.getMessages()	The raw messages array (or [])
res.getText()	Each message’s single text content block, joined with \n
res.body / res.getBody()	getText()’s output, JSON-parsed if it parses, else raw string, else null

Resource results

Member	Description
res.contents / res.getContents()	The raw contents array (or [])
res.getText()	Every text entry’s .text joined with \n (blob-only entries skipped)
res.body / res.getBody()	Same JSON-parse-or-raw-or-null behavior as tool/prompt, applied to getText()

Example: testing each kind

// Tool
test('sum is correct', () => {
  expect(res.getBody()).to.eql({ result: 4 })
})

// Prompt
test('renders a system + user turn', () => {
  expect(res.messages).to.have.lengthOf(2)
})

// Resource
test('returns non-empty text', () => {
  expect(res.getText().length).to.be.above(0)
})

Related

• Scripts run in an isolated sandbox with limits on memory and execution time, and no network or filesystem access — res is the only data a script can see.
• See Running Tests per Capability for how res gets its data — it’s always built from your most recent call to that capability.