Record tool usages in the Cline object, and persist them in the db for evals

evals/apps/cli/src/index.ts

@@ -275,7 +275,12 @@ const runExercise = async ({ run, task, server }: { run: Run; task: Task; server
 			})
 		}
 
-		if (eventName === RooCodeEventName.TaskCompleted || eventName === RooCodeEventName.TaskAborted) {
+		if (eventName === RooCodeEventName.TaskCompleted && taskMetricsId) {
+			const toolUsage = payload[2]
+			await updateTaskMetrics(taskMetricsId, { toolUsage })
+		}
+
+		if (eventName === RooCodeEventName.TaskAborted || eventName === RooCodeEventName.TaskCompleted) {
 			taskFinishedAt = Date.now()
 			await updateTask(task.id, { finishedAt: new Date() })
 		}

evals/packages/db/drizzle/0003_familiar_miss_america.sql

@@ -0,0 +1 @@
+ALTER TABLE `taskMetrics` ADD `toolUsage` blob;

evals/packages/db/drizzle/meta/0003_snapshot.json

@@ -0,0 +1,296 @@
+{
+	"version": "6",
+	"dialect": "sqlite",
+	"id": "a7a893e2-373a-4706-bcd4-772e2525db62",
+	"prevId": "f49d9b0b-fda9-467a-9adb-c941d6cbf7ce",
+	"tables": {
+		"runs": {
+			"name": "runs",
+			"columns": {
+				"id": {
+					"name": "id",
+					"type": "integer",
+					"primaryKey": true,
+					"notNull": true,
+					"autoincrement": true
+				},
+				"taskMetricsId": {
+					"name": "taskMetricsId",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": false,
+					"autoincrement": false
+				},
+				"model": {
+					"name": "model",
+					"type": "text",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				},
+				"description": {
+					"name": "description",
+					"type": "text",
+					"primaryKey": false,
+					"notNull": false,
+					"autoincrement": false
+				},
+				"settings": {
+					"name": "settings",
+					"type": "blob",
+					"primaryKey": false,
+					"notNull": false,
+					"autoincrement": false
+				},
+				"pid": {
+					"name": "pid",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": false,
+					"autoincrement": false
+				},
+				"socketPath": {
+					"name": "socketPath",
+					"type": "text",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				},
+				"concurrency": {
+					"name": "concurrency",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false,
+					"default": 2
+				},
+				"passed": {
+					"name": "passed",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false,
+					"default": 0
+				},
+				"failed": {
+					"name": "failed",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false,
+					"default": 0
+				},
+				"createdAt": {
+					"name": "createdAt",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				}
+			},
+			"indexes": {},
+			"foreignKeys": {
+				"runs_taskMetricsId_taskMetrics_id_fk": {
+					"name": "runs_taskMetricsId_taskMetrics_id_fk",
+					"tableFrom": "runs",
+					"tableTo": "taskMetrics",
+					"columnsFrom": ["taskMetricsId"],
+					"columnsTo": ["id"],
+					"onDelete": "no action",
+					"onUpdate": "no action"
+				}
+			},
+			"compositePrimaryKeys": {},
+			"uniqueConstraints": {},
+			"checkConstraints": {}
+		},
+		"taskMetrics": {
+			"name": "taskMetrics",
+			"columns": {
+				"id": {
+					"name": "id",
+					"type": "integer",
+					"primaryKey": true,
+					"notNull": true,
+					"autoincrement": true
+				},
+				"tokensIn": {
+					"name": "tokensIn",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				},
+				"tokensOut": {
+					"name": "tokensOut",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				},
+				"tokensContext": {
+					"name": "tokensContext",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				},
+				"cacheWrites": {
+					"name": "cacheWrites",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				},
+				"cacheReads": {
+					"name": "cacheReads",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				},
+				"cost": {
+					"name": "cost",
+					"type": "real",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				},
+				"duration": {
+					"name": "duration",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				},
+				"toolUsage": {
+					"name": "toolUsage",
+					"type": "blob",
+					"primaryKey": false,
+					"notNull": false,
+					"autoincrement": false
+				},
+				"createdAt": {
+					"name": "createdAt",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				}
+			},
+			"indexes": {},
+			"foreignKeys": {},
+			"compositePrimaryKeys": {},
+			"uniqueConstraints": {},
+			"checkConstraints": {}
+		},
+		"tasks": {
+			"name": "tasks",
+			"columns": {
+				"id": {
+					"name": "id",
+					"type": "integer",
+					"primaryKey": true,
+					"notNull": true,
+					"autoincrement": true
+				},
+				"runId": {
+					"name": "runId",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				},
+				"taskMetricsId": {
+					"name": "taskMetricsId",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": false,
+					"autoincrement": false
+				},
+				"language": {
+					"name": "language",
+					"type": "text",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				},
+				"exercise": {
+					"name": "exercise",
+					"type": "text",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				},
+				"passed": {
+					"name": "passed",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": false,
+					"autoincrement": false
+				},
+				"startedAt": {
+					"name": "startedAt",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": false,
+					"autoincrement": false
+				},
+				"finishedAt": {
+					"name": "finishedAt",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": false,
+					"autoincrement": false
+				},
+				"createdAt": {
+					"name": "createdAt",
+					"type": "integer",
+					"primaryKey": false,
+					"notNull": true,
+					"autoincrement": false
+				}
+			},
+			"indexes": {
+				"tasks_language_exercise_idx": {
+					"name": "tasks_language_exercise_idx",
+					"columns": ["runId", "language", "exercise"],
+					"isUnique": true
+				}
+			},
+			"foreignKeys": {
+				"tasks_runId_runs_id_fk": {
+					"name": "tasks_runId_runs_id_fk",
+					"tableFrom": "tasks",
+					"tableTo": "runs",
+					"columnsFrom": ["runId"],
+					"columnsTo": ["id"],
+					"onDelete": "no action",
+					"onUpdate": "no action"
+				},
+				"tasks_taskMetricsId_taskMetrics_id_fk": {
+					"name": "tasks_taskMetricsId_taskMetrics_id_fk",
+					"tableFrom": "tasks",
+					"tableTo": "taskMetrics",
+					"columnsFrom": ["taskMetricsId"],
+					"columnsTo": ["id"],
+					"onDelete": "no action",
+					"onUpdate": "no action"
+				}
+			},
+			"compositePrimaryKeys": {},
+			"uniqueConstraints": {},
+			"checkConstraints": {}
+		}
+	},
+	"views": {},
+	"enums": {},
+	"_meta": {
+		"schemas": {},
+		"tables": {},
+		"columns": {}
+	},
+	"internal": {
+		"indexes": {}
+	}
+}

evals/packages/db/drizzle/meta/_journal.json

@@ -22,6 +22,13 @@
 			"when": 1743698195142,
 			"tag": "0002_white_flatman",
 			"breakpoints": true
+		},
+		{
+			"idx": 3,
+			"version": "6",
+			"when": 1744933023667,
+			"tag": "0003_familiar_miss_america",
+			"breakpoints": true
 		}
 	]
 }

evals/packages/db/src/schema.ts

@@ -2,7 +2,7 @@ import { sqliteTable, text, real, integer, blob, uniqueIndex } from "drizzle-orm
 import { relations } from "drizzle-orm"
 import { createInsertSchema } from "drizzle-zod"
 
-import { RooCodeSettings, exerciseLanguages, rooCodeSettingsSchema } from "@evals/types"
+import { RooCodeSettings, ToolUsage, exerciseLanguages, rooCodeSettingsSchema, toolUsageSchema } from "@evals/types"
 
 /**
  * runs
@@ -84,12 +84,15 @@ export const taskMetrics = sqliteTable("taskMetrics", {
 	cacheReads: integer({ mode: "number" }).notNull(),
 	cost: real().notNull(),
 	duration: integer({ mode: "number" }).notNull(),
+	toolUsage: blob({ mode: "json" }).$type<ToolUsage>(),
 	createdAt: integer({ mode: "timestamp" }).notNull(),
 })
 
 export type TaskMetrics = typeof taskMetrics.$inferSelect
 
-export const insertTaskMetricsSchema = createInsertSchema(taskMetrics).omit({ id: true, createdAt: true })
+export const insertTaskMetricsSchema = createInsertSchema(taskMetrics)
+	.omit({ id: true, createdAt: true })
+	.extend({ toolUsage: toolUsageSchema.optional() })
 
 export type InsertTaskMetrics = Omit<typeof taskMetrics.$inferInsert, "id" | "createdAt">