perf(tools): read tool metadata instead of the registry on client paths

.agents/skills/tool-registry-boundary/SKILL.md

@@ -19,17 +19,20 @@ You keep the 4,300-tool executable registry out of module graphs that don't exec
 
 | you need | import | notes |
 | --- | --- | --- |
-| whether a tool id exists | `hasToolMetadata` from `@/tools/metadata` | |
-| a tool's params | `getToolParams` / `getToolMetadata` from `@/tools/metadata` | |
-| a tool's declared outputs | `getToolOutputsMetadata` from `@/tools/metadata-outputs` | separate module on purpose — see below |
-| every tool id | `getToolIds` from `@/tools/metadata` | |
+| whether a tool id exists | `hasToolId` from `@/tools/tool-ids` | ~110 KB — the cheapest module |
+| to resolve an unversioned name | `resolveToolId` from `@/tools/tool-ids` | |
+| every tool id | `getToolIds` from `@/tools/tool-ids` | |
+| a tool's params | `getToolParams` / `getToolMetadata` from `@/tools/metadata` | ~4 MB |
+| a tool's declared outputs | `getToolOutputsMetadata` from `@/tools/metadata-outputs` | ~4 MB, separate on purpose |
 | to **execute** a tool | `getTool` from `@/tools/utils`, or `@/tools/utils.server` | server paths only |
 
-Outputs live in their own module because they are roughly two thirds of the generated data and have a single consumer. Importing `@/tools/metadata` must never pull them — do not "helpfully" re-export one from the other.
+Three modules, cheapest first. Ids are their own artifact because resolution and existence checks need only the key set; outputs are their own because they are the larger half of the data with a single consumer. `@/tools/metadata` and `@/tools/metadata-outputs` both resolve ids through `@/tools/tool-ids`, which is what keeps them independent of each other — do not "helpfully" re-export one from another, or every caller pays for all three.
+
+All lookups guard with `Object.hasOwn`. `JSON.parse` yields an object with the normal prototype, so a bare bracket lookup returns inherited members: `getToolMetadata('constructor')` returned a *function* typed as tool metadata before that was fixed.
 
 ## The generated artifacts
 
-`apps/sim/tools/generated/tool-metadata.ts` and `tool-outputs.ts` are produced by `scripts/sync-tool-metadata.ts`:
+`apps/sim/tools/generated/tool-ids.ts`, `tool-metadata.ts` and `tool-outputs.ts` are produced by `scripts/sync-tool-metadata.ts`:
 
 ```bash
 bun run tool-metadata:generate   # after adding/changing a tool
@@ -43,6 +46,20 @@ Three non-obvious properties, each of which was measured and is easy to undo by
 - **The data is a JSON string parsed at runtime, not an imported `.json` and not an object literal.** With `resolveJsonModule` (which this repo enables), a `.json` import makes TypeScript infer a literal type for all 4,300+ entries and takes `tsc --noEmit` from **12.6s to 8m07s** — a 38x regression. An ambient `declare module` does *not* short-circuit it, and an object literal costs the same. A single string literal is one cheap token for both the compiler and the bundler, and `JSON.parse` beats evaluating the equivalent literal at runtime. Do not "clean this up" into a `.json` import.
 - **The generator refuses to emit function values.** If you add a field to `METADATA_FIELDS` that contains a closure, generation fails loudly rather than shipping executable config to the client. `hosting` and `schemaEnrichment` are excluded for exactly this reason (`hosting.enabled`, `pricing`, and `enrichSchema` are functions) — they are server-only.
 - **Empty param entries are stripped.** The registry contains one (`stt_deepgram_v2`), which crashes callers that read `param.type` while iterating.
+- **Lookups resolve versions.** `getTool` maps an unversioned name onto the newest version, and 246 tools are versioned. A plain key lookup would silently report them missing — a quiet correctness bug, not a crash. `resolveToolId` reproduces that against the id set and is differentially tested against the original.
+
+## Testing code that reads tool metadata
+
+Mock the module the code under test actually reads. `vi.mock('@/tools/utils', () => toolsUtilsMock)` only controls `getTool`; code that reads `params`/`outputs`/`name` goes through `@/tools/metadata`, so mocking `tools/utils` there is a **no-op that still passes** — because the real generated artifacts happen to agree with the mock fixtures. The test looks green while controlling nothing.
+
+```ts
+import { blocksMock, toolsMetadataMock, toolsUtilsMock } from '@sim/testing/mocks'
+
+vi.mock('@/tools/utils', () => toolsUtilsMock)      // executable lookup
+vi.mock('@/tools/metadata', () => toolsMetadataMock) // params / outputs / name
+```
+
+Both are backed by the same `mockToolConfigs`, so mocking both gives one consistent tool universe. If you are unsure whether a mock is load-bearing, change a fixture value to a sentinel and confirm the test fails.
 
 ## How to verify an edge actually got cut
 
@@ -59,9 +76,9 @@ Reference points measured on this repo:
 | `tools/registry.ts` reachable | ~4,900 |
 | `tools/merge-params.ts` (leaf) | 2 |
 | `providers/utils.ts` after cutting its `params` edge | 22 |
-| `app/workspace/[workspaceId]/w/page.tsx` (canvas) | 6,591, of which 4,689 are the registry |
+| `app/workspace/[workspaceId]/w/page.tsx` (canvas) | 6,592 before, 1,908 after |
 
-The canvas route reaches the registry through **four** redundant edges — `providers/utils` (via `tools/params`), `lib/workflows/blocks/block-outputs`, `lib/workflows/sanitization/validation`, and `serializer/index`. Cutting any one alone moves the module count by ~1. They must all be cut before anything improves; measure the route, not the file you edited.
+The canvas route reached the registry through **four** redundant edges — `providers/utils` (via `tools/params`), `lib/workflows/blocks/block-outputs`, `lib/workflows/sanitization/validation`, and `serializer/index`. Cutting any one alone moved the module count by ~1. They all had to go before anything improved; measure the route, not the file you edited.
 
 ## When adding a new caller
 

.claude/commands/tool-registry-boundary.md

@@ -18,17 +18,20 @@ You keep the 4,300-tool executable registry out of module graphs that don't exec
 
 | you need | import | notes |
 | --- | --- | --- |
-| whether a tool id exists | `hasToolMetadata` from `@/tools/metadata` | |
-| a tool's params | `getToolParams` / `getToolMetadata` from `@/tools/metadata` | |
-| a tool's declared outputs | `getToolOutputsMetadata` from `@/tools/metadata-outputs` | separate module on purpose — see below |
-| every tool id | `getToolIds` from `@/tools/metadata` | |
+| whether a tool id exists | `hasToolId` from `@/tools/tool-ids` | ~110 KB — the cheapest module |
+| to resolve an unversioned name | `resolveToolId` from `@/tools/tool-ids` | |
+| every tool id | `getToolIds` from `@/tools/tool-ids` | |
+| a tool's params | `getToolParams` / `getToolMetadata` from `@/tools/metadata` | ~4 MB |
+| a tool's declared outputs | `getToolOutputsMetadata` from `@/tools/metadata-outputs` | ~4 MB, separate on purpose |
 | to **execute** a tool | `getTool` from `@/tools/utils`, or `@/tools/utils.server` | server paths only |
 
-Outputs live in their own module because they are roughly two thirds of the generated data and have a single consumer. Importing `@/tools/metadata` must never pull them — do not "helpfully" re-export one from the other.
+Three modules, cheapest first. Ids are their own artifact because resolution and existence checks need only the key set; outputs are their own because they are the larger half of the data with a single consumer. `@/tools/metadata` and `@/tools/metadata-outputs` both resolve ids through `@/tools/tool-ids`, which is what keeps them independent of each other — do not "helpfully" re-export one from another, or every caller pays for all three.
+
+All lookups guard with `Object.hasOwn`. `JSON.parse` yields an object with the normal prototype, so a bare bracket lookup returns inherited members: `getToolMetadata('constructor')` returned a *function* typed as tool metadata before that was fixed.
 
 ## The generated artifacts
 
-`apps/sim/tools/generated/tool-metadata.ts` and `tool-outputs.ts` are produced by `scripts/sync-tool-metadata.ts`:
+`apps/sim/tools/generated/tool-ids.ts`, `tool-metadata.ts` and `tool-outputs.ts` are produced by `scripts/sync-tool-metadata.ts`:
 
 ```bash
 bun run tool-metadata:generate   # after adding/changing a tool
@@ -42,6 +45,20 @@ Three non-obvious properties, each of which was measured and is easy to undo by
 - **The data is a JSON string parsed at runtime, not an imported `.json` and not an object literal.** With `resolveJsonModule` (which this repo enables), a `.json` import makes TypeScript infer a literal type for all 4,300+ entries and takes `tsc --noEmit` from **12.6s to 8m07s** — a 38x regression. An ambient `declare module` does *not* short-circuit it, and an object literal costs the same. A single string literal is one cheap token for both the compiler and the bundler, and `JSON.parse` beats evaluating the equivalent literal at runtime. Do not "clean this up" into a `.json` import.
 - **The generator refuses to emit function values.** If you add a field to `METADATA_FIELDS` that contains a closure, generation fails loudly rather than shipping executable config to the client. `hosting` and `schemaEnrichment` are excluded for exactly this reason (`hosting.enabled`, `pricing`, and `enrichSchema` are functions) — they are server-only.
 - **Empty param entries are stripped.** The registry contains one (`stt_deepgram_v2`), which crashes callers that read `param.type` while iterating.
+- **Lookups resolve versions.** `getTool` maps an unversioned name onto the newest version, and 246 tools are versioned. A plain key lookup would silently report them missing — a quiet correctness bug, not a crash. `resolveToolId` reproduces that against the id set and is differentially tested against the original.
+
+## Testing code that reads tool metadata
+
+Mock the module the code under test actually reads. `vi.mock('@/tools/utils', () => toolsUtilsMock)` only controls `getTool`; code that reads `params`/`outputs`/`name` goes through `@/tools/metadata`, so mocking `tools/utils` there is a **no-op that still passes** — because the real generated artifacts happen to agree with the mock fixtures. The test looks green while controlling nothing.
+
+```ts
+import { blocksMock, toolsMetadataMock, toolsUtilsMock } from '@sim/testing/mocks'
+
+vi.mock('@/tools/utils', () => toolsUtilsMock)      // executable lookup
+vi.mock('@/tools/metadata', () => toolsMetadataMock) // params / outputs / name
+```
+
+Both are backed by the same `mockToolConfigs`, so mocking both gives one consistent tool universe. If you are unsure whether a mock is load-bearing, change a fixture value to a sentinel and confirm the test fails.
 
 ## How to verify an edge actually got cut
 
@@ -58,9 +75,9 @@ Reference points measured on this repo:
 | `tools/registry.ts` reachable | ~4,900 |
 | `tools/merge-params.ts` (leaf) | 2 |
 | `providers/utils.ts` after cutting its `params` edge | 22 |
-| `app/workspace/[workspaceId]/w/page.tsx` (canvas) | 6,591, of which 4,689 are the registry |
+| `app/workspace/[workspaceId]/w/page.tsx` (canvas) | 6,592 before, 1,908 after |
 
-The canvas route reaches the registry through **four** redundant edges — `providers/utils` (via `tools/params`), `lib/workflows/blocks/block-outputs`, `lib/workflows/sanitization/validation`, and `serializer/index`. Cutting any one alone moves the module count by ~1. They must all be cut before anything improves; measure the route, not the file you edited.
+The canvas route reached the registry through **four** redundant edges — `providers/utils` (via `tools/params`), `lib/workflows/blocks/block-outputs`, `lib/workflows/sanitization/validation`, and `serializer/index`. Cutting any one alone moved the module count by ~1. They all had to go before anything improved; measure the route, not the file you edited.
 
 ## When adding a new caller
 

.cursor/commands/tool-registry-boundary.md

@@ -14,17 +14,20 @@ You keep the 4,300-tool executable registry out of module graphs that don't exec
 
 | you need | import | notes |
 | --- | --- | --- |
-| whether a tool id exists | `hasToolMetadata` from `@/tools/metadata` | |
-| a tool's params | `getToolParams` / `getToolMetadata` from `@/tools/metadata` | |
-| a tool's declared outputs | `getToolOutputsMetadata` from `@/tools/metadata-outputs` | separate module on purpose — see below |
-| every tool id | `getToolIds` from `@/tools/metadata` | |
+| whether a tool id exists | `hasToolId` from `@/tools/tool-ids` | ~110 KB — the cheapest module |
+| to resolve an unversioned name | `resolveToolId` from `@/tools/tool-ids` | |
+| every tool id | `getToolIds` from `@/tools/tool-ids` | |
+| a tool's params | `getToolParams` / `getToolMetadata` from `@/tools/metadata` | ~4 MB |
+| a tool's declared outputs | `getToolOutputsMetadata` from `@/tools/metadata-outputs` | ~4 MB, separate on purpose |
 | to **execute** a tool | `getTool` from `@/tools/utils`, or `@/tools/utils.server` | server paths only |
 
-Outputs live in their own module because they are roughly two thirds of the generated data and have a single consumer. Importing `@/tools/metadata` must never pull them — do not "helpfully" re-export one from the other.
+Three modules, cheapest first. Ids are their own artifact because resolution and existence checks need only the key set; outputs are their own because they are the larger half of the data with a single consumer. `@/tools/metadata` and `@/tools/metadata-outputs` both resolve ids through `@/tools/tool-ids`, which is what keeps them independent of each other — do not "helpfully" re-export one from another, or every caller pays for all three.
+
+All lookups guard with `Object.hasOwn`. `JSON.parse` yields an object with the normal prototype, so a bare bracket lookup returns inherited members: `getToolMetadata('constructor')` returned a *function* typed as tool metadata before that was fixed.
 
 ## The generated artifacts
 
-`apps/sim/tools/generated/tool-metadata.ts` and `tool-outputs.ts` are produced by `scripts/sync-tool-metadata.ts`:
+`apps/sim/tools/generated/tool-ids.ts`, `tool-metadata.ts` and `tool-outputs.ts` are produced by `scripts/sync-tool-metadata.ts`:
 
 ```bash
 bun run tool-metadata:generate   # after adding/changing a tool
@@ -38,6 +41,20 @@ Three non-obvious properties, each of which was measured and is easy to undo by
 - **The data is a JSON string parsed at runtime, not an imported `.json` and not an object literal.** With `resolveJsonModule` (which this repo enables), a `.json` import makes TypeScript infer a literal type for all 4,300+ entries and takes `tsc --noEmit` from **12.6s to 8m07s** — a 38x regression. An ambient `declare module` does *not* short-circuit it, and an object literal costs the same. A single string literal is one cheap token for both the compiler and the bundler, and `JSON.parse` beats evaluating the equivalent literal at runtime. Do not "clean this up" into a `.json` import.
 - **The generator refuses to emit function values.** If you add a field to `METADATA_FIELDS` that contains a closure, generation fails loudly rather than shipping executable config to the client. `hosting` and `schemaEnrichment` are excluded for exactly this reason (`hosting.enabled`, `pricing`, and `enrichSchema` are functions) — they are server-only.
 - **Empty param entries are stripped.** The registry contains one (`stt_deepgram_v2`), which crashes callers that read `param.type` while iterating.
+- **Lookups resolve versions.** `getTool` maps an unversioned name onto the newest version, and 246 tools are versioned. A plain key lookup would silently report them missing — a quiet correctness bug, not a crash. `resolveToolId` reproduces that against the id set and is differentially tested against the original.
+
+## Testing code that reads tool metadata
+
+Mock the module the code under test actually reads. `vi.mock('@/tools/utils', () => toolsUtilsMock)` only controls `getTool`; code that reads `params`/`outputs`/`name` goes through `@/tools/metadata`, so mocking `tools/utils` there is a **no-op that still passes** — because the real generated artifacts happen to agree with the mock fixtures. The test looks green while controlling nothing.
+
+```ts
+import { blocksMock, toolsMetadataMock, toolsUtilsMock } from '@sim/testing/mocks'
+
+vi.mock('@/tools/utils', () => toolsUtilsMock)      // executable lookup
+vi.mock('@/tools/metadata', () => toolsMetadataMock) // params / outputs / name
+```
+
+Both are backed by the same `mockToolConfigs`, so mocking both gives one consistent tool universe. If you are unsure whether a mock is load-bearing, change a fixture value to a sentinel and confirm the test fails.
 
 ## How to verify an edge actually got cut
 
@@ -54,9 +71,9 @@ Reference points measured on this repo:
 | `tools/registry.ts` reachable | ~4,900 |
 | `tools/merge-params.ts` (leaf) | 2 |
 | `providers/utils.ts` after cutting its `params` edge | 22 |
-| `app/workspace/[workspaceId]/w/page.tsx` (canvas) | 6,591, of which 4,689 are the registry |
+| `app/workspace/[workspaceId]/w/page.tsx` (canvas) | 6,592 before, 1,908 after |
 
-The canvas route reaches the registry through **four** redundant edges — `providers/utils` (via `tools/params`), `lib/workflows/blocks/block-outputs`, `lib/workflows/sanitization/validation`, and `serializer/index`. Cutting any one alone moves the module count by ~1. They must all be cut before anything improves; measure the route, not the file you edited.
+The canvas route reached the registry through **four** redundant edges — `providers/utils` (via `tools/params`), `lib/workflows/blocks/block-outputs`, `lib/workflows/sanitization/validation`, and `serializer/index`. Cutting any one alone moved the module count by ~1. They all had to go before anything improved; measure the route, not the file you edited.
 
 ## When adding a new caller
 

apps/sim/ee/access-control/components/group-detail.tsx

@@ -76,7 +76,7 @@ import {
 import type { ProviderId } from '@/providers/types'
 import { getAllProviderIds, getProviderFromModel } from '@/providers/utils'
 import type { ProviderName } from '@/stores/providers'
-import { getTool } from '@/tools/utils'
+import { getToolMetadata } from '@/tools/metadata'
 
 const logger = createLogger('AccessControlGroupDetail')
 
@@ -733,7 +733,7 @@ function BlockToolRow({
   const checkboxId = `block-${block.type}`
 
   const toolItems = useMemo<DenylistGridItem[]>(
-    () => (block.tools?.access ?? []).map((id) => ({ id, label: getTool(id)?.name ?? id })),
+    () => (block.tools?.access ?? []).map((id) => ({ id, label: getToolMetadata(id)?.name ?? id })),
     [block.tools?.access]
   )
   const isExpandable = toolItems.length > 1

apps/sim/lib/workflows/blocks/block-outputs.ts

@@ -22,7 +22,7 @@ import {
   type OutputCondition,
   type OutputFieldDefinition,
 } from '@/blocks/types'
-import { getTool } from '@/tools/utils'
+import { getToolOutputsMetadata } from '@/tools/metadata-outputs'
 import { getTrigger, isTriggerValid } from '@/triggers'
 
 const logger = createLogger('BlockOutputs')
@@ -681,13 +681,13 @@ export function getToolOutputs(
     const toolId = blockConfig.tools.config.tool(params)
     if (!toolId) return {}
 
-    const toolConfig = getTool(toolId)
-    if (!toolConfig?.outputs) return {}
+    const toolOutputs = getToolOutputsMetadata(toolId)
+    if (!toolOutputs) return {}
     if (includeHidden) {
-      return toolConfig.outputs
+      return toolOutputs
     }
     return Object.fromEntries(
-      Object.entries(toolConfig.outputs).filter(([_, def]) => !isHiddenFromDisplay(def))
+      Object.entries(toolOutputs).filter(([_, def]) => !isHiddenFromDisplay(def))
     )
   } catch (error) {
     logger.warn('Failed to get tool outputs', { error })

apps/sim/lib/workflows/sanitization/validation.ts

@@ -4,7 +4,7 @@ import { isRecordLike } from '@sim/utils/object'
 import { getBlock } from '@/blocks/registry'
 import { isCustomTool, isMcpTool } from '@/executor/constants'
 import type { BlockState, WorkflowState } from '@/stores/workflows/workflow/types'
-import { getTool } from '@/tools/utils'
+import { hasToolId } from '@/tools/tool-ids'
 
 const logger = createLogger('WorkflowValidation')
 
@@ -305,8 +305,7 @@ export function validateToolReference(
 
   if (!isCustomTool(toolId) && !isMcpTool(toolId)) {
     // For built-in tools, verify they exist
-    const tool = getTool(toolId)
-    if (!tool) {
+    if (!hasToolId(toolId)) {
       return `Block ${blockName || 'unknown'} (${blockType}): references non-existent tool '${toolId}'`
     }
   }