@alpic-ai/tracks 0.16.0 → 0.18.0

package/dist/main.js

@@ -475,7 +475,7 @@ const FailureKind = Schema.Literals([
 	"unexpected"
 ]).annotate({
 	identifier: "FailureKind",
-	description: "What kind of thing went wrong: the repository could not be checked out, the environment's setup command failed, the browser did not start, the sandbox disappeared, the agent's process was lost, the sandbox provider failed, the agent of a run that accepts no follow-ups stopped with an error, or something unexpected."
+	description: "What kind of thing went wrong: the repository could not be checked out, the environment's setup command failed, the browser did not start, the sandbox disappeared, the agent's process was lost, the sandbox provider failed, the agent stopped with an error on the last prompt of a one-shot run or of a run you ended, or something unexpected."
 });
 const runFields = {
 	id: RunId,
@@ -567,17 +567,15 @@ var RunNotIdle = class extends Schema.TaggedError()("RunNotIdle", { message: Sch
 }) {};
 const Count$1 = Schema.Natural;
 const RunUsage = Schema.Struct({
-	cost_usd: Schema.Number.annotate({ description: "Total cost of the run so far, in US dollars." }),
+	cost_usd: Schema.Number.annotate({ description: "What was taken from your balance for the run so far, its model calls and its run fee, in US dollars." }),
+	run_fee_usd: Schema.Number.annotate({ description: "The run fee, in US dollars: charged once the run has ended, unless it failed on our side, 0 until then." }),
 	models: Schema.Array(Schema.Struct({
 		provider: Schema.String.annotate({ description: "Company serving the model, such as `anthropic`." }),
 		model: Schema.String,
 		quantities: Schema.Record(Schema.String, Count$1).annotate({ description: "Units consumed, keyed by unit as the provider counts them, such as input tokens or web searches." }),
 		cost_usd: Schema.Number.annotate({ description: "Cost of this model's usage, in US dollars." })
 	})).annotate({ description: "Model usage, one entry per model." }),
-	sandbox: Schema.Struct({
-		seconds: Schema.Number.annotate({ description: "Time the sandbox ran, in seconds." }),
-		cost_usd: Schema.Number.annotate({ description: "Cost of the sandbox time, in US dollars." })
-	})
+	sandbox: Schema.Struct({ seconds: Schema.Number.annotate({ description: "Time the sandbox ran, in seconds, covered by the run fee." }) })
 }).annotate({
 	identifier: "RunUsage",
 	description: "What a run consumed and what it cost."
@@ -765,14 +763,14 @@ var RunsApi = class extends HttpApiGroup.make("runs").add(HttpApiEndpoint.post("
 	error: RunNotFound
 }).annotateMerge(OpenApi.annotations({
 	summary: "Get a run's usage",
-	description: "Returns what the run has consumed so far, per model and in sandbox time, and what it cost."
+	description: "Returns what the run has consumed so far, per model and in sandbox time, and what it is charged."
 })), HttpApiEndpoint.post("end", "/runs/:id/end", {
 	params: { id: RunId },
 	success: Run.pipe(HttpApiSchema.status(202)),
 	error: RunNotFound
 }).annotateMerge(OpenApi.annotations({
 	summary: "End a run",
-	description: "Ends the run with `finished` once its current prompt finishes, then destroys its sandbox. Ending a run that has ended does nothing."
+	description: "Ends the run once its current prompt finishes, with `finished`, or `errored` if the agent stopped with an error on that prompt, then destroys its sandbox. Ending a run that has ended does nothing."
 })), HttpApiEndpoint.post("cancel", "/runs/:id/cancel", {
 	params: { id: RunId },
 	success: Run.pipe(HttpApiSchema.status(202)),
@@ -918,6 +916,16 @@ const ExcludedVariant = Schema.Struct({
 	identifier: "ExcludedVariant",
 	description: "A combination of repository and agent tools not to run."
 });
+const Persona = Schema.Literals([
+	"none",
+	"default",
+	"developer",
+	"vibe_coder",
+	"product_manager"
+]).annotate({
+	identifier: "Persona",
+	description: "Who answers the agent when it stops, up to 5 times a run, so each run is a conversation: `none` (no one, the agent gets the prompt alone), `default` (a plain user who tells it to go ahead), `developer` (a terse technical user who expects idiomatic, tested work), `vibe_coder` (a non-technical user happy with whatever works) or `product_manager` (a user focused on users and outcomes, who answers requirements and leaves technical choices to the agent). Its replies are written from the scenario's prompt, never from its evaluations, and are free. A conversation takes more time and money than one prompt: raise the time limit and budget. Defaults to `default` for a new experiment; omitted from an update, the experiment keeps its own."
+});
 const RunSettings = Schema.Struct({
 	repetitions: Schema.Int.check(Schema.isGreaterThanOrEqualTo(1)).annotate({ description: "Identical runs started for each scenario, variant and agent, to compare results beyond chance." }),
 	time_limit_seconds: Schema.Finite.check(Schema.isBetween({
@@ -930,6 +938,7 @@ const RunSettings = Schema.Struct({
 		exclusiveMinimum: true,
 		maximum: MAX_BUDGET_USD
 	})).annotate({ description: `Most each run may spend, in US dollars, above 0 and at most ${MAX_BUDGET_USD}.` }),
+	persona: Persona,
 	setup: script("Shell script run in each run's sandbox after checkout and before the prompt."),
 	teardown: script("Shell script run in each run's sandbox after the run ends."),
 	variables: Schema.Array(Schema.Struct({
@@ -961,7 +970,10 @@ const ScenarioInput = Scenario.mapFields(Struct.evolve({
 	input_files: (inputFiles) => inputFiles.pipe(Schema.withDecodingDefaultKey(Effect.succeed([]))),
 	evaluations: () => withUniqueIds(EvaluationInput, "evaluations")
 })).annotate({ identifier: "ScenarioInput" });
-const RunSettingsInput = RunSettings.mapFields(Struct.evolve({ variables: (variables) => variables.pipe(Schema.withDecodingDefaultKey(Effect.succeed([]))) })).annotate({ identifier: "RunSettingsInput" });
+const RunSettingsInput = RunSettings.mapFields(Struct.evolve({
+	persona: (persona) => Schema.optionalKey(persona),
+	variables: (variables) => variables.pipe(Schema.withDecodingDefaultKey(Effect.succeed([])))
+})).annotate({ identifier: "RunSettingsInput" });
 const unknownExclusions = (input) => input.excluded_variants.flatMap((excluded, index) => [...input.agent_tools.some((agentTools) => agentTools.id === excluded.agent_tools_id) ? [] : [{
 	path: [
 		"excluded_variants",
@@ -1364,7 +1376,7 @@ const isDatabaseVariable = (name) => DATABASE_ENV.some((databaseName) => databas
 SERVICE_ENV.flatMap((name) => isSecret(name) && !isDatabaseVariable(name) ? [name] : []);
 const readEnv = (...names) => Config.schema(Env.mapFields(Struct.pick(names)));
 // package.json
-var version = "0.16.0";
+var version = "0.18.0";
 // src/connections.ts
 const encodeConnection = Schema.encodeEffect(Schema.fromJsonString(Connection, { space: 2 }));
 const encodeConnections = Schema.encodeEffect(Schema.fromJsonString(List(Connection), { space: 2 }));

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@alpic-ai/tracks",
-  "version": "0.16.0",
+  "version": "0.18.0",
   "description": "Run coding agents in sandboxes from the command line",
   "license": "UNLICENSED",
   "repository": {