Files
supabase/apps/studio/components/ui/AIAssistantPanel/Message.Parts.tsx
Matt Rossman d143571586 feat(assistant): trace-level scorers + server-side tool execution with needsApproval (#45654)
## Motivation

When Assistant runs a potentially destructive tool like `execute_sql`,
it stops the LLM request and prompts for client-side approval and
execution of the tool. After approval, a second request kicks off under
a separate trace. This has made scoring and
[Topics](https://www.braintrust.dev/blog/topics) classification
challenging, as the generated `output` is split across stateless
requests. The [span-level
scoring](https://www.braintrust.dev/docs/evaluate/custom-code#score-spans)
approach we've used thusfar (after the LLM call, we massage the result
into an `output` payload that's stuck onto the root span) has been
cumbersome and led to invalid scores / topics where only part of the
assistant response is considered. It's also inefficient, as we're
duplicating potentially large info (like the `search_docs` output) that
already exists within the trace.

An alternative to scoring spans is to [score
traces](https://www.braintrust.dev/docs/evaluate/custom-code#score-traces).
Braintrust [best
practices](https://www.braintrust.dev/docs/evaluate/score-online#best-practices)
advise:

> Use span scope for evaluating individual operations or outputs. Use
trace scope for evaluating multi-turn conversations, overall workflow
completion, or when your scorer needs access to the full execution
context.

We've also received [direct
guidance](https://supabase.slack.com/archives/C05QYJBLX89/p1777925770927149?thread_ts=1777905716.911979&cid=C05QYJBLX89)
from their team to use this approach.

## Changes

Migrates eval scorers from custom `AssistantEvalOutput` shape to
trace-level scoring via `trace.getThread()` / `trace.getSpans()`, with
thread parsing that scores the full latest Assistant turn and passes
prior conversation separately where relevant.

Moves `execute_sql` and `deploy_edge_function` from client-side
execution after approval to AI SDK `needsApproval` + server-side
`execute()`. SQL results returned to the model are gated by AI opt-in
level, so row data is only included with `schema_and_log_and_data`;
otherwise the tool returns the no-data-permissions sentinel.

Adds `metadata.isFinalStep` to disambiguate multiple LLM requests within
an "assistant" turn due to tool call requests/responses. For online
evals, this means we should configure automations to only score traces
with `metadata.isFinalStep = true` to ensure we're judging the complete
generated response.

Other minor kaizen changes:
- Renamed `promptProviderOptions` to `systemProviderOptions` to clarify
that this is associated with the "system" message and disambiguate from
the root `providerOptions`
- Adds `evals/trace-utils.ts` to handle Zod validation of the `unknown`
span shapes from Braintrust, to more easily access typed inputs/output
on tool spans.
- Bumps AI SDK floor version `^6.0.116` → `^6.0.174`
- Tweaked the "Conciseness" scorer to not unfairly dock points for the
new `[called tool_name]` labels in serialized assistant response

## Verification

In the studio staging build, I asked Assistant to create a todos table
with 3 sample todos. I manually approved the `execute_sql` call and saw
Assistant generate text before & after the call.

In Braintrust I verified two traces were produced (see [filtered
logs](https://www.braintrust.dev/app/supabase.io/p/Assistant/logs?v=Staging&tvt=trace&search={%22filter%22:[{%22text%22:%22metadata.environment%2520%253D%2520%27staging%27%22,%22label%22:%22metadata.environment%2520%253D%2520%27staging%27%22,%22originType%22:%22btql%22},{%22text%22:%22%2560Chat%2520ID%2560%2520%253D%2520%25221cb2ac45-e5e7-458c-9da4-3bf6863b8842%2522%22,%22label%22:%22Chat%2520ID%2520equals%25201cb2ac45-e5e7-458c-9da4-3bf6863b8842%22,%22originType%22:%22form%22}]})),
the first with `metadata.isFinalStep = false` and the second with
`metadata.isFinalStep = true`.

In the Braintrust staging scorers, I ran the preview Completeness scorer
on the second trace and verified it sees the complete Assistant response
including markers for tool calls ([link to
trace](https://www.braintrust.dev/app/supabase.io/p/Assistant%20(Staging%20Scorers)/trace?object_type=project_logs&object_id=b5214b62-ad1e-4929-9d5b-40b1daebe948&r=0ed0a4f8-8aff-4a34-bb1d-1df1d88a5070&s=ff9015f8-6bf7-4ab3-83a9-ca4e69e27e82))

<img width="1193" height="960" alt="CleanShot 2026-05-07 at 11 27 10@2x"
src="https://github.com/user-attachments/assets/509d4858-c3a1-4068-986d-3aa4d5617d1a"
/>

I also tested the `deploy_edge_function` workflow and verified it still
prompts for permission and warns on deployment of existing functions.

**References**
- https://www.braintrust.dev/docs/evaluate/custom-code#score-traces
-
https://ai-sdk.dev/docs/ai-sdk-core/tools-and-tool-calling#tool-execution-approval

Supercedes https://github.com/supabase/supabase/pull/45556 and
https://github.com/supabase/supabase/pull/45339

Closes AI-473

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

* **New Features**
* Tool actions (SQL execution, edge-function deploy) now require
explicit user Approve/Deny before proceeding.

* **Improvements**
* Assistant pauses for approval responses before sending follow-ups,
giving clearer control over risky actions.
  * Deploy/replace flows show confirmation and clearer replace warnings.
* Evaluation/scoring updated to use richer trace data for more accurate
assistant performance signals.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
2026-05-12 15:24:21 -04:00

274 lines
8.3 KiB
TypeScript

import { UIMessage as VercelMessage } from '@ai-sdk/react'
import { type DynamicToolUIPart, type ReasoningUIPart, type TextUIPart, type ToolUIPart } from 'ai'
import { BrainIcon, CheckIcon, Loader2 } from 'lucide-react'
import { cn } from 'ui'
import { DisplayBlockRenderer } from './DisplayBlockRenderer'
import { EdgeFunctionRenderer } from './EdgeFunctionRenderer'
import { Tool } from './elements/Tool'
import { useMessageActionsContext, useMessageInfoContext } from './Message.Context'
import {
deployEdgeFunctionInputSchema,
deployEdgeFunctionOutputSchema,
parseExecuteSqlChartResult,
} from './Message.utils'
import { MessageMarkdown } from './MessageMarkdown'
function MessagePartText({ textPart }: { textPart: TextUIPart }) {
const { id, isLoading, readOnly, isUserMessage, state } = useMessageInfoContext()
return (
<MessageMarkdown
id={id}
isLoading={isLoading}
readOnly={readOnly}
className={cn(
'max-w-none space-y-4 prose prose-sm prose-li:mt-1 [&>div]:my-4 prose-h1:text-xl prose-h1:mt-6 prose-h2:text-lg prose-h2:font-medium prose-h3:no-underline prose-h3:text-base prose-h3:mb-4 prose-strong:font-medium prose-strong:text-foreground prose-ol:space-y-3 prose-ul:space-y-3 prose-li:my-0 wrap-break-word [&>p:not(:last-child)]:mb-2! [&>*>p:first-child]:mt-0! [&>*>p:last-child]:mb-0! [&>*>*>p:first-child]:mt-0! [&>*>*>p:last-child]:mb-0! [&>ol>li]:pl-4!',
isUserMessage && 'text-foreground [&>p]:font-medium',
state === 'editing' && 'animate-pulse'
)}
>
{textPart.text}
</MessageMarkdown>
)
}
function MessagePartDynamicTool({ toolPart }: { toolPart: DynamicToolUIPart }) {
return (
<Tool
icon={
toolPart.state === 'input-streaming' ? (
<Loader2 strokeWidth={1.5} size={12} className="animate-spin" />
) : (
<CheckIcon strokeWidth={1.5} size={12} className="text-foreground-muted" />
)
}
label={
<div>
{toolPart.state === 'input-streaming' ? 'Running ' : 'Ran '}
<span className="text-foreground-lighter">{`${toolPart.toolName}`}</span>
</div>
}
/>
)
}
function MessagePartTool({ toolPart }: { toolPart: ToolUIPart }) {
return (
<Tool
icon={
toolPart.state === 'input-streaming' ? (
<Loader2 strokeWidth={1.5} size={12} className="animate-spin" />
) : (
<CheckIcon strokeWidth={1.5} size={12} className="text-foreground-muted" />
)
}
label={
<div>
{toolPart.state === 'input-streaming' ? 'Running ' : 'Ran '}
<span className="text-foreground-lighter">{`${toolPart.type.replace('tool-', '')}`}</span>
</div>
}
/>
)
}
function MessagePartReasoning({ reasoningPart }: { reasoningPart: ReasoningUIPart }) {
return (
<Tool
icon={
reasoningPart.state === 'streaming' ? (
<Loader2 strokeWidth={1.5} size={12} className="animate-spin" />
) : (
<BrainIcon strokeWidth={1.5} size={12} className="text-foreground-muted" />
)
}
label={reasoningPart.state === 'streaming' ? 'Thinking...' : 'Reasoned'}
>
{reasoningPart.text}
</Tool>
)
}
function ToolDisplayExecuteSqlLoading({ label = 'Writing SQL...' }: { label?: string }) {
return (
<div className="my-4 rounded-lg border bg-surface-75 heading-meta h-9 px-3 text-foreground-light flex items-center gap-2">
<Loader2 className="w-4 h-4 animate-spin" />
{label}
</div>
)
}
function ToolDisplayExecuteSqlFailure() {
return <div className="text-xs text-danger">Failed to execute SQL.</div>
}
function MessagePartExecuteSql({
toolPart,
isLastPart,
}: {
toolPart: ToolUIPart
isLastPart?: boolean
}) {
const { id, isLastMessage } = useMessageInfoContext()
const { addToolApprovalResponse } = useMessageActionsContext()
const { toolCallId, state, input, output } = toolPart
if (state === 'input-streaming') {
return <ToolDisplayExecuteSqlLoading />
}
if (state === 'output-error') {
return <ToolDisplayExecuteSqlFailure />
}
if (state === 'approval-responded') {
return <ToolDisplayExecuteSqlLoading label="Running SQL..." />
}
const { data: chart, success } = parseExecuteSqlChartResult(input)
if (!success) return null
if (
state === 'input-available' ||
state === 'approval-requested' ||
state === 'output-available'
) {
const approvalId = state === 'approval-requested' ? toolPart.approval?.id : undefined
return (
<div className="w-auto overflow-x-hidden my-4 space-y-2">
<DisplayBlockRenderer
messageId={id}
toolCallId={toolCallId}
initialArgs={{
sql: chart.sql,
label: chart.label,
isWriteQuery: chart.isWriteQuery,
view: chart.view,
xAxis: chart.xAxis,
yAxis: chart.yAxis,
}}
initialResults={output}
toolState={state}
isLastPart={isLastPart}
isLastMessage={isLastMessage}
onApprove={
approvalId
? () => addToolApprovalResponse?.({ id: approvalId, approved: true })
: undefined
}
onDeny={
approvalId
? () => addToolApprovalResponse?.({ id: approvalId, approved: false })
: undefined
}
/>
</div>
)
}
return null
}
const TOOL_DEPLOY_EDGE_FUNCTION_STATES_WITH_INPUT = new Set([
'input-available',
'approval-requested',
'approval-responded',
'output-available',
])
function MessagePartDeployEdgeFunction({ toolPart }: { toolPart: ToolUIPart }) {
const { state, input, output } = toolPart
const { addToolApprovalResponse } = useMessageActionsContext()
if (state === 'input-streaming') {
return (
<div className="my-4 rounded-lg border bg-surface-75 heading-meta h-9 px-3 text-foreground-light flex items-center gap-2">
<Loader2 className="w-4 h-4 animate-spin" />
Writing Edge Function...
</div>
)
}
if (state === 'output-error') {
return <p className="text-xs text-danger">Failed to deploy Edge Function.</p>
}
if (!TOOL_DEPLOY_EDGE_FUNCTION_STATES_WITH_INPUT.has(state)) return null
const parsedInput = deployEdgeFunctionInputSchema.safeParse(input)
if (!parsedInput.success) return null
const parsedOutput = deployEdgeFunctionOutputSchema.safeParse(output)
const isInitiallyDeployed =
state === 'output-available' && parsedOutput.success && parsedOutput.data.success === true
const approvalId = state === 'approval-requested' ? toolPart.approval?.id : undefined
return (
<EdgeFunctionRenderer
label={parsedInput.data.label}
code={parsedInput.data.code}
functionName={parsedInput.data.functionName}
showConfirmFooter={state === 'approval-requested'}
isDeploying={state === 'approval-responded'}
initialIsDeployed={isInitiallyDeployed}
onApprove={
approvalId ? () => addToolApprovalResponse?.({ id: approvalId, approved: true }) : undefined
}
onDeny={
approvalId
? () => addToolApprovalResponse?.({ id: approvalId, approved: false })
: undefined
}
/>
)
}
const MessagePart = {
Text: MessagePartText,
Dynamic: MessagePartDynamicTool,
Tool: MessagePartTool,
Reasoning: MessagePartReasoning,
ExecuteSql: MessagePartExecuteSql,
DeployEdgeFunction: MessagePartDeployEdgeFunction,
} as const
export function MessagePartSwitcher({
part,
isLastPart,
}: {
part: NonNullable<VercelMessage['parts']>[number]
isLastPart?: boolean
}) {
switch (part.type) {
case 'dynamic-tool': {
return <MessagePart.Dynamic toolPart={part} />
}
case 'tool-list_policies':
case 'tool-search_docs':
case 'tool-get_active_incidents':
case 'tool-load_knowledge': {
return <MessagePart.Tool toolPart={part} />
}
case 'reasoning':
return <MessagePart.Reasoning reasoningPart={part} />
case 'text':
return <MessagePart.Text textPart={part} />
case 'tool-execute_sql': {
return <MessagePart.ExecuteSql toolPart={part} isLastPart={isLastPart} />
}
case 'tool-deploy_edge_function': {
return <MessagePart.DeployEdgeFunction toolPart={part} />
}
case 'source-url':
case 'source-document':
case 'file':
default:
return null
}
}