mirror of
https://github.com/supabase/supabase.git
synced 2026-10-19 16:19:10 +03:00
## 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 -->
330 lines
9.4 KiB
TypeScript
330 lines
9.4 KiB
TypeScript
import assert from 'node:assert'
|
|
import { tool } from 'ai'
|
|
import { z } from 'zod'
|
|
|
|
import { getStudioTools } from '../tools/studio-tools'
|
|
import { getMcpTools } from '@/lib/ai/tools/mcp-tools'
|
|
|
|
const listTablesInputSchema = z.object({
|
|
schemas: z.array(z.string()).describe('The schema names to list.'),
|
|
})
|
|
|
|
const getAdvisorsInputSchema = z.object({
|
|
type: z.enum(['security', 'performance']).optional(),
|
|
})
|
|
|
|
const getLogsInputSchema = z.object({
|
|
limit: z.number().min(1).max(100).optional(),
|
|
level: z.enum(['debug', 'info', 'warning', 'error']).optional(),
|
|
source: z.enum(['postgres', 'auth', 'storage', 'edge_function']).optional(),
|
|
search: z.string().optional(),
|
|
})
|
|
|
|
const listPoliciesInputSchema = z.object({
|
|
schemas: z.array(z.string()).describe('The schema names to get the policies for'),
|
|
})
|
|
|
|
export const MOCK_TABLES_DATA = [
|
|
{
|
|
name: 'user_documents',
|
|
rls_enabled: false,
|
|
columns: [
|
|
{ name: 'id', data_type: 'bigint' },
|
|
{ name: 'user_id', data_type: 'uuid' },
|
|
{ name: 'title', data_type: 'text' },
|
|
],
|
|
},
|
|
{
|
|
name: 'customers',
|
|
rls_enabled: true,
|
|
columns: [
|
|
{ name: 'id', data_type: 'uuid' },
|
|
{ name: 'tenant_id', data_type: 'uuid' },
|
|
{ name: 'email', data_type: 'text' },
|
|
],
|
|
},
|
|
{
|
|
name: 'projects',
|
|
rls_enabled: false,
|
|
columns: [
|
|
{ name: 'id', data_type: 'uuid' },
|
|
{ name: 'organization_id', data_type: 'uuid' },
|
|
{ name: 'name', data_type: 'text' },
|
|
],
|
|
},
|
|
{
|
|
name: 'user_organizations',
|
|
rls_enabled: true,
|
|
columns: [
|
|
{ name: 'user_id', data_type: 'uuid' },
|
|
{ name: 'organization_id', data_type: 'uuid' },
|
|
],
|
|
},
|
|
]
|
|
|
|
const MOCK_EXTENSIONS_DATA = [
|
|
{ name: 'pgcrypto', schema: 'extensions', installed_version: '1.3' },
|
|
{ name: 'uuid-ossp', schema: 'extensions', installed_version: '1.1' },
|
|
]
|
|
|
|
const MOCK_EDGE_FUNCTIONS_DATA = [
|
|
{ name: 'hello-world', last_deployed_at: '2024-06-10T12:30:00Z' },
|
|
{ name: 'daily-metrics-sync', last_deployed_at: '2024-06-18T08:15:00Z' },
|
|
{ name: 'select-from-table-with-auth-rls', last_deployed_at: '2024-06-19T09:20:00Z' },
|
|
]
|
|
|
|
const MOCK_ADVISORIES_DATA = [
|
|
{
|
|
id: '0016_materialized_view_in_api',
|
|
level: 'warning',
|
|
category: 'security',
|
|
message: 'Materialized views in API schema can bypass RLS. Move them to private schema.',
|
|
remediationUrl:
|
|
'https://supabase.com/docs/guides/database/database-advisors?queryGroups=lint&lint=0016_materialized_view_in_api',
|
|
},
|
|
{
|
|
id: '0031_functions_no_rls_guard',
|
|
level: 'notice',
|
|
category: 'security',
|
|
message: 'Function api.health_check should verify auth context before querying tables.',
|
|
remediationUrl:
|
|
'https://supabase.com/docs/guides/database/database-advisors?queryGroups=lint&lint=0031_functions_no_rls_guard',
|
|
},
|
|
{
|
|
id: '1012_slow_query',
|
|
level: 'info',
|
|
category: 'performance',
|
|
message:
|
|
'Query on table edge_function_logs exceeded 3s average execution time over the last hour.',
|
|
remediationUrl: 'https://supabase.com/docs/guides/platform/performance-advisors#slow-queries',
|
|
},
|
|
]
|
|
|
|
const MOCK_LOGS_DATA = [
|
|
{
|
|
id: 'log-001',
|
|
timestamp: '2024-06-20T14:12:00Z',
|
|
level: 'error',
|
|
source: 'edge_function' as const,
|
|
target: 'hello-world',
|
|
message: "TypeError: fetch failed at await supabase.functions.invoke('analytics')",
|
|
},
|
|
{
|
|
id: 'log-002',
|
|
timestamp: '2024-06-20T14:05:30Z',
|
|
level: 'warning',
|
|
source: 'postgres' as const,
|
|
target: 'connection_pool',
|
|
message: 'Query timeout exceeded for statement SELECT * FROM public.audit_log_entries',
|
|
},
|
|
{
|
|
id: 'log-003',
|
|
timestamp: '2024-06-20T13:59:10Z',
|
|
level: 'info',
|
|
source: 'edge_function' as const,
|
|
target: 'daily-metrics-sync',
|
|
message: 'Invocation completed in 520ms',
|
|
},
|
|
{
|
|
id: 'log-004',
|
|
timestamp: '2024-06-20T13:50:00Z',
|
|
level: 'error',
|
|
source: 'postgres' as const,
|
|
target: 'trigger:refresh_materialized_views',
|
|
message: 'permission denied for relation user_documents',
|
|
},
|
|
{
|
|
id: 'log-005',
|
|
timestamp: '2024-06-20T13:45:00Z',
|
|
level: 'info',
|
|
source: 'auth' as const,
|
|
target: 'email-confirmation',
|
|
message: 'Sent verification email to alex@example.com',
|
|
},
|
|
]
|
|
|
|
function createMockedStudioTools() {
|
|
const studioTools = getStudioTools()
|
|
|
|
return Object.fromEntries(
|
|
Object.entries(studioTools).map(([name, baseTool]) => {
|
|
// Always mock execute_sql and deploy_edge_function with needsApproval disabled
|
|
if (name === 'execute_sql') {
|
|
return [name, { ...baseTool, needsApproval: false, execute: async () => [] as unknown[] }]
|
|
}
|
|
if (name === 'deploy_edge_function') {
|
|
return [
|
|
name,
|
|
{ ...baseTool, needsApproval: false, execute: async () => ({ success: true }) },
|
|
]
|
|
}
|
|
if (typeof baseTool.execute === 'function') {
|
|
return [name, baseTool]
|
|
}
|
|
|
|
return [
|
|
name,
|
|
{ ...baseTool, execute: async () => ({ status: 'Tool call mocked successfully.' }) },
|
|
]
|
|
})
|
|
) as typeof studioTools
|
|
}
|
|
|
|
function createMockListTablesTool(overrideData?: Record<string, typeof MOCK_TABLES_DATA>) {
|
|
return tool({
|
|
description: 'Lists tables and columns for the provided schemas.',
|
|
inputSchema: listTablesInputSchema,
|
|
execute: async ({ schemas }: { schemas: string[] }) => {
|
|
const effectiveSchemas = schemas?.length ? schemas : ['public']
|
|
return effectiveSchemas.map((schema) => ({
|
|
schema,
|
|
tables: overrideData?.[schema] ?? MOCK_TABLES_DATA,
|
|
}))
|
|
},
|
|
})
|
|
}
|
|
|
|
function createMockListExtensionsTool() {
|
|
return tool({
|
|
description: 'Lists installed database extensions.',
|
|
inputSchema: z.object({}),
|
|
execute: async () => {
|
|
return MOCK_EXTENSIONS_DATA
|
|
},
|
|
})
|
|
}
|
|
|
|
function createMockListEdgeFunctionsTool() {
|
|
return tool({
|
|
description: 'Lists available Supabase Edge Functions.',
|
|
inputSchema: z.object({}),
|
|
execute: async () => {
|
|
return MOCK_EDGE_FUNCTIONS_DATA
|
|
},
|
|
})
|
|
}
|
|
|
|
function createMockGetAdvisorsTool() {
|
|
return tool({
|
|
description: 'Returns advisory notices for the project (mocked).',
|
|
inputSchema: getAdvisorsInputSchema,
|
|
execute: async ({ type }: { type?: 'security' | 'performance' }) => {
|
|
if (type) {
|
|
return MOCK_ADVISORIES_DATA.filter((advisory) => advisory.category === type)
|
|
}
|
|
return MOCK_ADVISORIES_DATA
|
|
},
|
|
})
|
|
}
|
|
|
|
function createMockGetLogsTool() {
|
|
return tool({
|
|
description: 'Fetches recent project logs for debugging or health checks (mocked).',
|
|
inputSchema: getLogsInputSchema,
|
|
execute: async ({
|
|
limit = 10,
|
|
level,
|
|
source,
|
|
search,
|
|
}: {
|
|
limit?: number
|
|
level?: 'debug' | 'info' | 'warning' | 'error'
|
|
source?: 'postgres' | 'auth' | 'storage' | 'edge_function'
|
|
search?: string
|
|
}) => {
|
|
let filtered = MOCK_LOGS_DATA
|
|
|
|
if (level) {
|
|
filtered = filtered.filter((entry) => entry.level === level)
|
|
}
|
|
|
|
if (source) {
|
|
filtered = filtered.filter((entry) => entry.source === source)
|
|
}
|
|
|
|
if (search) {
|
|
const needle = search.toLowerCase()
|
|
filtered = filtered.filter((entry) =>
|
|
`${entry.message} ${entry.target}`.toLowerCase().includes(needle)
|
|
)
|
|
}
|
|
|
|
return filtered.slice(0, limit)
|
|
},
|
|
})
|
|
}
|
|
|
|
function createMockListPoliciesTool() {
|
|
return tool({
|
|
description: 'Get existing RLS policies for provided schemas.',
|
|
inputSchema: listPoliciesInputSchema,
|
|
execute: async ({ schemas }: { schemas: string[] }) => {
|
|
const effectiveSchemas = schemas?.length ? schemas : ['public']
|
|
const results = [] as Array<{
|
|
schema: string
|
|
table: string
|
|
policies: Array<{
|
|
name: string
|
|
command: 'select' | 'insert' | 'update' | 'delete'
|
|
using?: string
|
|
check?: string
|
|
}>
|
|
}>
|
|
|
|
for (const schema of effectiveSchemas) {
|
|
if (schema !== 'public') continue
|
|
results.push(
|
|
{
|
|
schema,
|
|
table: 'customers',
|
|
policies: [
|
|
{
|
|
name: 'customers_tenant_select',
|
|
command: 'select',
|
|
using: "(auth.jwt() ->> 'tenant_id')::uuid = tenant_id",
|
|
},
|
|
],
|
|
},
|
|
{ schema, table: 'user_documents', policies: [] },
|
|
{ schema, table: 'projects', policies: [] }
|
|
)
|
|
}
|
|
return results
|
|
},
|
|
})
|
|
}
|
|
|
|
export type MockToolOverrides = {
|
|
list_tables?: Record<string, typeof MOCK_TABLES_DATA>
|
|
}
|
|
|
|
/**
|
|
* Deterministic mock implementations of MCP/platform tools for evals.
|
|
* These mirror tool names used in prompts so the model can call them,
|
|
* but return stable, static data for repeatable tests.
|
|
*
|
|
* Note: search_docs uses the real implementation
|
|
*/
|
|
export async function getMockTools(overrides?: MockToolOverrides) {
|
|
const mockedStudioTools = createMockedStudioTools()
|
|
|
|
const { search_docs } = await getMcpTools({
|
|
accessToken: 'mock-access-token',
|
|
projectRef: 'mock-project-ref',
|
|
aiOptInLevel: 'schema_and_log_and_data',
|
|
})
|
|
|
|
assert(search_docs, 'search_docs tool not available from MCP server')
|
|
|
|
return {
|
|
...mockedStudioTools,
|
|
search_docs,
|
|
list_tables: createMockListTablesTool(overrides?.list_tables),
|
|
list_extensions: createMockListExtensionsTool(),
|
|
list_edge_functions: createMockListEdgeFunctionsTool(),
|
|
get_advisors: createMockGetAdvisorsTool(),
|
|
get_logs: createMockGetLogsTool(),
|
|
list_policies: createMockListPoliciesTool(),
|
|
}
|
|
}
|