anpord 0.1.11 → 0.1.13

package/dist/cli-runtime.cjs

@@ -118,8 +118,8 @@ const executeCli = (definition, args, journal = JOURNAL) => effect.Effect.gen(fu
 		stdout: ""
 	}))));
 }).pipe(effect.Effect.withSpan("CliMock.execute"), effect.Effect.runPromise);
-const runCli = async (definition) => {
-	const result = await executeCli(definition, process.argv.slice(2));
+const runCli = async (definition, journal) => {
+	const result = await executeCli(definition, process.argv.slice(2), journal);
 	process.stdout.write(result.stdout);
 	process.stderr.write(result.stderr);
 	process.exitCode = result.exitCode;

package/dist/cli-runtime.d.cts

@@ -6,6 +6,6 @@ interface CliExecution {
   readonly stdout: string;
 }
 declare const executeCli: (definition: CliDefinition, args: readonly string[], journal?: string) => Promise<CliExecution>;
-declare const runCli: (definition: CliDefinition) => Promise<void>;
+declare const runCli: (definition: CliDefinition, journal?: string) => Promise<void>;
 //#endregion
 export { CliExecution, executeCli, runCli };

package/dist/cli-runtime.d.mts

@@ -6,6 +6,6 @@ interface CliExecution {
   readonly stdout: string;
 }
 declare const executeCli: (definition: CliDefinition, args: readonly string[], journal?: string) => Promise<CliExecution>;
-declare const runCli: (definition: CliDefinition) => Promise<void>;
+declare const runCli: (definition: CliDefinition, journal?: string) => Promise<void>;
 //#endregion
 export { CliExecution, executeCli, runCli };

package/dist/cli-runtime.mjs

@@ -117,8 +117,8 @@ const executeCli = (definition, args, journal = JOURNAL) => Effect.gen(function*
 		stdout: ""
 	}))));
 }).pipe(Effect.withSpan("CliMock.execute"), Effect.runPromise);
-const runCli = async (definition) => {
-	const result = await executeCli(definition, process.argv.slice(2));
+const runCli = async (definition, journal) => {
+	const result = await executeCli(definition, process.argv.slice(2), journal);
 	process.stdout.write(result.stdout);
 	process.stderr.write(result.stderr);
 	process.exitCode = result.exitCode;

package/dist/client-BfdjFtlV.mjs

@@ -0,0 +1,305 @@
+import { C as EvalTrigger, E as EvalVerify, T as EvalPrompt, _ as HarnessProfile, a as EvalPageCursor, c as EvalRunPage, d as EvalValidator, f as EvalVariables, h as StartedEval, i as EvalName, l as EvalSandbox, m as RerunCellRequest, n as EvalCellHistoryEntry, o as EvalPrepare, p as ModelCatalogue, r as EvalHarness, s as EvalRun, t as CaseCache, u as EvalSource, v as PROFILE_HARNESS_RULE, w as EvalCaseName, x as profileFitsHarness } from "./evals-Dcc_x6jW.mjs";
+import { FetchHttpClient, HttpApi, HttpApiClient, HttpApiEndpoint, HttpApiGroup, HttpApiMiddleware, HttpApiSchema, HttpApiSecurity, HttpClient, HttpClientRequest, OpenApi } from "@effect/platform";
+import { Context, Effect, Layer, Redacted, Schema } from "effect";
+//#region ../schema/src/domain/prompts.ts
+const ChannelName = Schema.String.pipe(Schema.minLength(1), Schema.maxLength(36), Schema.pattern(/^[a-z0-9][a-z0-9_-]*$/, { message: () => "Channel must be lowercase alphanumeric, optionally with - or _" }), Schema.brand("ChannelName")).annotations({ description: "Addresses a version. `latest` is derived from the highest version rather than stored, so it cannot drift from the version table." });
+ChannelName.make("production");
+ChannelName.make("latest");
+const PromptId = Schema.String.pipe(Schema.minLength(1), Schema.maxLength(255), Schema.pattern(/^[a-z0-9][a-z0-9/_-]*$/, { message: () => "Prompt id must be lowercase alphanumeric, optionally with / _ or -" }), Schema.brand("PromptId"));
+const PromptName = Schema.String.pipe(Schema.minLength(1), Schema.maxLength(255), Schema.brand("PromptName"));
+const VersionNumber = Schema.Int.pipe(Schema.positive(), Schema.brand("VersionNumber"));
+Schema.NumberFromString.pipe(Schema.int(), Schema.positive(), Schema.brand("VersionNumber"));
+Schema.NumberFromString.pipe(Schema.int(), Schema.positive(), Schema.lessThanOrEqualTo(100));
+const Timestamp = Schema.Union(Schema.DateFromSelf, Schema.Date);
+const CommitMessage = Schema.String.pipe(Schema.maxLength(500));
+const PromptContent = Schema.String.pipe(Schema.minLength(1), Schema.maxLength(262144, { message: () => "Prompt content is longer than 256 KB" }));
+const PromptDescription = Schema.String.pipe(Schema.maxLength(2e3));
+const PromptConfig = Schema.Record({
+	key: Schema.String,
+	value: Schema.Unknown
+});
+const Author = Schema.Struct({
+	image: Schema.NullOr(Schema.String),
+	name: Schema.String
+});
+Schema.Struct({
+	author: Schema.NullOr(Author),
+	channel: Schema.NullOr(ChannelName),
+	commitMessage: Schema.NullOr(CommitMessage),
+	config: PromptConfig,
+	content: Schema.String,
+	createdAt: Timestamp,
+	id: PromptId,
+	name: PromptName,
+	version: VersionNumber,
+	versionId: Schema.String
+});
+Schema.Struct({
+	channel: ChannelName,
+	updatedAt: Timestamp,
+	updatedBy: Schema.NullOr(Author),
+	version: VersionNumber
+});
+const PromptSummary = Schema.Struct({
+	author: Schema.NullOr(Author),
+	description: Schema.NullOr(Schema.String),
+	id: PromptId,
+	latestVersion: Schema.NullOr(VersionNumber),
+	name: PromptName,
+	productionVersion: Schema.NullOr(VersionNumber),
+	updatedAt: Timestamp
+});
+Schema.Struct({
+	id: PromptId,
+	updatedAt: Timestamp
+});
+Schema.Literal("all", "draft", "live");
+Schema.Literal("name", "updated");
+Schema.Struct({
+	items: Schema.Array(PromptSummary),
+	nextCursor: Schema.NullOr(Schema.String)
+});
+Schema.Struct({
+	commitMessage: Schema.optional(CommitMessage),
+	config: Schema.optional(PromptConfig),
+	content: PromptContent,
+	description: Schema.optional(PromptDescription),
+	id: PromptId,
+	name: PromptName,
+	publish: Schema.optional(Schema.Boolean)
+});
+Schema.Struct({
+	commitMessage: Schema.optional(CommitMessage),
+	config: Schema.optional(PromptConfig),
+	content: PromptContent,
+	publish: Schema.optional(Schema.Boolean)
+});
+Schema.Struct({
+	commitMessage: Schema.optional(CommitMessage),
+	config: Schema.optional(PromptConfig),
+	content: PromptContent
+});
+Schema.Struct({
+	description: Schema.optional(PromptDescription),
+	id: Schema.optional(PromptId),
+	name: Schema.optional(PromptName)
+});
+Schema.Struct({
+	channel: ChannelName,
+	version: VersionNumber
+});
+Schema.Struct({
+	channel: Schema.optional(ChannelName),
+	version: Schema.optional(VersionNumber)
+});
+//#endregion
+//#region ../schema/src/domain/errors.ts
+var NotFound = class extends Schema.TaggedError()("NotFound", { message: Schema.String }, HttpApiSchema.annotations({ status: 404 })) {};
+var Conflict = class extends Schema.TaggedError()("Conflict", { message: Schema.String }, HttpApiSchema.annotations({ status: 409 })) {};
+var BadRequest = class extends Schema.TaggedError()("BadRequest", { message: Schema.String }, HttpApiSchema.annotations({ status: 400 })) {};
+var Unauthorized = class extends Schema.TaggedError()("Unauthorized", { message: Schema.String }, HttpApiSchema.annotations({ status: 401 })) {};
+var Forbidden = class extends Schema.TaggedError()("Forbidden", { message: Schema.String }, HttpApiSchema.annotations({ status: 403 })) {};
+Schema.TaggedError()("InternalError", { message: Schema.String }, HttpApiSchema.annotations({ status: 500 }));
+//#endregion
+//#region ../schema/src/internal/authentication.ts
+var CurrentActor = class extends Context.Tag("@anpord/schema/CurrentActor")() {};
+HttpApiMiddleware.Tag()("@anpord/schema/Authentication", {
+	failure: Unauthorized,
+	provides: CurrentActor,
+	security: { session: HttpApiSecurity.apiKey({
+		in: "cookie",
+		key: "anpord.session_token"
+	}) }
+});
+//#endregion
+//#region ../schema/src/public/authentication.ts
+var ApiKeyAuthentication = class extends HttpApiMiddleware.Tag()("@anpord/schema/ApiKeyAuthentication", {
+	failure: Unauthorized,
+	provides: CurrentActor,
+	security: { bearer: HttpApiSecurity.bearer }
+}) {};
+//#endregion
+//#region ../schema/src/public/evals-api.ts
+const EvalRunRequest = Schema.Struct({ id: Schema.String }).annotations({
+	description: "Select an eval run by id.",
+	identifier: "EvalRunRequest"
+});
+const EvalCellRequest = Schema.Struct({ cellKey: Schema.String }).annotations({
+	description: "Select an eval cell by its stable key.",
+	identifier: "EvalCellRequest"
+});
+const EvalModelsRequest = Schema.Struct({
+	harness: EvalHarness,
+	q: Schema.optional(Schema.String)
+}).annotations({
+	description: "Select a harness whose available models should be listed.",
+	identifier: "EvalModelsRequest"
+});
+const PublicEvalSandbox = EvalSandbox.annotations({
+	description: "The hosted sandbox a task runs in.",
+	identifier: "PublicEvalSandbox"
+});
+const ListEvalsRequest = Schema.Struct({
+	cursor: Schema.optional(Schema.NullOr(EvalPageCursor)),
+	limit: Schema.optional(Schema.Int)
+}).annotations({
+	description: "Where to read from, and how much.",
+	identifier: "ListEvalsRequest"
+});
+const PublicEvalCase = Schema.Struct({
+	cache: Schema.optional(CaseCache),
+	name: EvalCaseName,
+	prepare: Schema.optional(Schema.NullOr(EvalPrepare)),
+	source: Schema.optional(EvalSource),
+	validator: Schema.optional(Schema.NullOr(EvalValidator)),
+	variables: Schema.optional(EvalVariables),
+	verify: Schema.NullOr(EvalVerify)
+}).pipe(Schema.filter(({ validator, verify }) => !(validator !== void 0 && validator !== null && verify !== null), { message: () => "Use either validator or verify, not both." })).annotations({
+	description: "A task, workspace source, setup command, and verifier.",
+	identifier: "StartEvalCase"
+});
+const PublicEvalTask = Schema.Struct({
+	harness: EvalHarness,
+	model: Schema.String.pipe(Schema.minLength(1)),
+	profile: Schema.optional(HarnessProfile),
+	sandbox: Schema.optional(PublicEvalSandbox)
+}).pipe(Schema.filter(profileFitsHarness, { message: () => PROFILE_HARNESS_RULE })).annotations({
+	description: `A harness and model, with an optional sandbox and an optional profile layered on the harness. Omit the sandbox to use the default. ${PROFILE_HARNESS_RULE}`,
+	identifier: "StartEvalTask"
+});
+const PublicStartEvalRequest = Schema.Struct({
+	trigger: Schema.optional(EvalTrigger),
+	cases: Schema.Array(PublicEvalCase).pipe(Schema.minItems(1), Schema.maxItems(100)),
+	name: Schema.optional(EvalName),
+	prompt: EvalPrompt,
+	tasks: Schema.Array(PublicEvalTask).pipe(Schema.minItems(1), Schema.maxItems(20)),
+	trials: Schema.Int.pipe(Schema.between(1, 10))
+}).annotations({
+	description: `Start a grid with at most 100 total case, task, and trial combinations.`,
+	identifier: "StartEvalRequest"
+});
+var PublicEvalsGroup = class extends HttpApiGroup.make("evals").add(HttpApiEndpoint.post("list", "/evals.list").setPayload(ListEvalsRequest).addSuccess(EvalRunPage).annotate(OpenApi.Summary, "List eval runs").annotate(OpenApi.Description, "Newest first. Pass the `next` cursor from a response to read the page after it; a null `next` means there are no more.")).add(HttpApiEndpoint.post("start", "/evals.start").setPayload(PublicStartEvalRequest).addSuccess(StartedEval).annotate(OpenApi.Summary, "Start an eval run").annotate(OpenApi.Description, `Starts the grid and returns its id while trials continue in the background. ${PROFILE_HARNESS_RULE}`)).add(HttpApiEndpoint.post("get", "/evals.get").setPayload(EvalRunRequest).addSuccess(EvalRun).annotate(OpenApi.Summary, "Get an eval run")).add(HttpApiEndpoint.post("cellHistory", "/evals.cellHistory").setPayload(EvalCellRequest).addSuccess(Schema.Array(EvalCellHistoryEntry)).annotate(OpenApi.Summary, "List a cell's history").annotate(OpenApi.Description, "Returns the 20 most recent results.")).add(HttpApiEndpoint.post("rerunCell", "/evals.rerunCell").setPayload(Schema.extend(EvalRunRequest, Schema.extend(EvalCellRequest, RerunCellRequest))).addSuccess(StartedEval).annotate(OpenApi.Summary, "Rerun one cell")).add(HttpApiEndpoint.post("models", "/evals.models").setPayload(EvalModelsRequest).addSuccess(ModelCatalogue).annotate(OpenApi.Summary, "List models available to the harness").annotate(OpenApi.Description, "The command harness has no catalogue of its own, so its list is empty: the model is whatever the profile's run command reads from ANPORD_MODEL.")).addError(BadRequest).addError(Conflict).addError(Forbidden).addError(NotFound).middleware(ApiKeyAuthentication).annotate(OpenApi.Title, "Evals").annotate(OpenApi.Description, "Run cases across harness, model, and sandbox combinations and compare the results with their baselines.") {};
+//#endregion
+//#region ../schema/src/public/requests.ts
+const GetPromptRequest = Schema.Struct({
+	channel: Schema.optional(ChannelName),
+	id: PromptId,
+	includeVersions: Schema.optional(Schema.Boolean),
+	version: Schema.optional(VersionNumber)
+}).annotations({
+	description: "Resolve a prompt. Give a version to pin one, a channel to follow one, or neither for the organization's default channel.",
+	identifier: "GetPromptRequest"
+});
+Schema.Struct({ id: PromptId }).annotations({
+	description: "Show a prompt's version history.",
+	identifier: "ListVersionsRequest"
+});
+const ListPromptsRequest = Schema.Record({
+	key: Schema.String,
+	value: Schema.Never
+}).annotations({
+	description: "No parameters; returns up to 100 prompts in the organization.",
+	identifier: "ListPromptsRequest"
+});
+const CreatePromptRequest = Schema.Struct({
+	config: Schema.optional(PromptConfig),
+	content: PromptContent,
+	description: Schema.optional(Schema.String),
+	id: PromptId,
+	message: Schema.optional(CommitMessage),
+	name: PromptName
+}).annotations({
+	description: "Create a prompt and publish its first version to the default channel.",
+	identifier: "CreatePromptRequest"
+});
+const UpdatePromptRequest = Schema.Struct({
+	config: Schema.optional(PromptConfig),
+	content: PromptContent,
+	id: PromptId,
+	message: Schema.optional(CommitMessage)
+}).annotations({
+	description: "Add a version to a prompt. The new version becomes latest, and omitted config is stored as an empty object.",
+	identifier: "UpdatePromptRequest"
+});
+const PromotePromptRequest = Schema.Struct({
+	channel: ChannelName,
+	id: PromptId,
+	version: VersionNumber
+}).annotations({
+	description: "Point a channel at a version.",
+	identifier: "PromotePromptRequest"
+});
+Schema.Struct({ id: PromptId }).annotations({
+	description: "Archive a prompt. Existing selectors stay readable.",
+	identifier: "ArchivePromptRequest"
+});
+const Ok = Schema.Struct({ ok: Schema.Literal(true) }).annotations({
+	description: "The operation succeeded.",
+	identifier: "Ok"
+});
+//#endregion
+//#region ../schema/src/public/shapes.ts
+const Instant = Schema.DateTimeUtc.annotations({
+	description: "An ISO-8601 timestamp in UTC.",
+	identifier: "Instant",
+	jsonSchema: { format: "date-time" }
+});
+const PublicPrompt = Schema.Struct({
+	channel: Schema.NullOr(ChannelName),
+	config: PromptConfig,
+	content: Schema.String,
+	createdAt: Instant,
+	id: PromptId,
+	message: Schema.NullOr(CommitMessage),
+	name: PromptName,
+	version: VersionNumber
+}).annotations({
+	description: "A prompt resolved at a specific version.",
+	identifier: "ResolvedPrompt"
+});
+const PublicVersion = Schema.Struct({
+	createdAt: Instant,
+	message: Schema.NullOr(CommitMessage),
+	version: VersionNumber
+}).annotations({
+	description: "One entry in a prompt's history.",
+	identifier: "Version"
+});
+const PublicPromptWithVersions = Schema.extend(PublicPrompt, Schema.Struct({ versions: Schema.optional(Schema.Array(PublicVersion)) })).annotations({
+	description: "A prompt, with its version history when `includeVersions` is set.",
+	identifier: "Prompt"
+});
+const PublicPromptSummary = Schema.Struct({
+	id: PromptId,
+	latestVersion: Schema.NullOr(VersionNumber),
+	name: PromptName,
+	productionVersion: Schema.NullOr(VersionNumber),
+	updatedAt: Instant
+}).annotations({
+	description: "A prompt without its content.",
+	identifier: "PromptSummary"
+});
+const PromptList = Schema.Struct({ data: Schema.Array(PublicPromptSummary) }).annotations({
+	description: "Every prompt in the organization.",
+	identifier: "PromptList"
+});
+//#endregion
+//#region ../schema/src/public/prompts-api.ts
+var PublicPromptsGroup = class extends HttpApiGroup.make("prompts").add(HttpApiEndpoint.post("get", "/prompts.get").setPayload(GetPromptRequest).addSuccess(PublicPromptWithVersions).annotate(OpenApi.Summary, "Resolve a prompt").annotate(OpenApi.Description, "Returns the content a caller should send to a model. With no selector this follows the organization's default channel.")).add(HttpApiEndpoint.post("list", "/prompts.list").setPayload(ListPromptsRequest).addSuccess(PromptList).annotate(OpenApi.Summary, "List prompts").annotate(OpenApi.Description, "Up to 100 prompts in the organization, without content.")).add(HttpApiEndpoint.post("create", "/prompts.create").setPayload(CreatePromptRequest).addSuccess(PublicPromptWithVersions).annotate(OpenApi.Summary, "Create a prompt").annotate(OpenApi.Description, "Creates the prompt and its first version in one call.")).add(HttpApiEndpoint.post("update", "/prompts.update").setPayload(UpdatePromptRequest).addSuccess(PublicPromptWithVersions).annotate(OpenApi.Summary, "Add a version").annotate(OpenApi.Description, "Content is versioned, so updating a prompt appends a version rather than overwriting one. Earlier versions stay readable.")).add(HttpApiEndpoint.post("promote", "/prompts.promote").setPayload(PromotePromptRequest).addSuccess(Ok).annotate(OpenApi.Summary, "Promote a version to a channel").annotate(OpenApi.Description, "Points a channel, such as production, at a version. This is how a version goes live without callers changing anything.")).addError(BadRequest).addError(Forbidden).addError(Conflict).addError(NotFound).middleware(ApiKeyAuthentication).annotate(OpenApi.Title, "Prompts").annotate(OpenApi.Description, "Read and write prompts. Content is versioned, so writes append rather than overwrite, and channels decide which version callers receive.") {};
+//#endregion
+//#region ../schema/src/public/api.ts
+var PublicApi = class extends HttpApi.make("anpord-public").add(PublicEvalsGroup).add(PublicPromptsGroup).prefix("/v1").annotate(OpenApi.Title, "Anpord API").annotate(OpenApi.Version, "1.0.0").annotate(OpenApi.Description, "Run agent evals and manage prompts. Every endpoint takes a JSON body over POST and authenticates with a bearer API key.").annotate(OpenApi.Servers, [{
+	description: "Production",
+	url: "https://api.anpord.com"
+}]) {};
+//#endregion
+//#region ../schema/src/public/client.ts
+const DEFAULT_BASE_URL = "https://api.anpord.com";
+const make = ({ apiKey, baseUrl = DEFAULT_BASE_URL }) => HttpApiClient.make(PublicApi, {
+	baseUrl,
+	transformClient: HttpClient.mapRequest(HttpClientRequest.bearerToken(Redacted.value(apiKey)))
+});
+var AnpordApi = class extends Effect.Tag("@anpord/sdk/AnpordApi")() {};
+const layer = (options) => Layer.effect(AnpordApi, make(options)).pipe(Layer.provide(FetchHttpClient.layer));
+//#endregion
+export { ChannelName as a, make as i, DEFAULT_BASE_URL as n, PromptId as o, layer as r, VersionNumber as s, AnpordApi as t };