@ory/argus 0.9.1 → 0.11.0

package/dist/testing.js

@@ -0,0 +1,372 @@
+"use strict";
+/**
+ * Shared test utilities for Ory agent plugin integration tests.
+ *
+ * These helpers provide mock Ory API responses, error factories, and
+ * common assertion patterns so that harness plugin tests focus on
+ * harness-specific behavior rather than duplicating boilerplate.
+ */
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.BATCH_SERVER_DENIED = exports.BATCH_SERVER_ALLOWED_TOOL_DENIED = exports.BATCH_BOTH_ALLOWED = exports.PERMISSION_DENIED = exports.PERMISSION_ALLOWED = exports.MOCK_INACTIVE_OAUTH2_RESPONSE = exports.MOCK_OAUTH2_RESPONSE = exports.MOCK_INACTIVE_SESSION_RESPONSE = exports.MOCK_SESSION_RESPONSE = exports.runHarnessContractSuite = void 0;
+exports.createMockClient = createMockClient;
+exports.stubApi = stubApi;
+exports.makeAxiosError = makeAxiosError;
+exports.makeNetworkError = makeNetworkError;
+exports.makeRateLimitError = makeRateLimitError;
+exports.makeMfaRequiredError = makeMfaRequiredError;
+exports.makeSessionInactiveError = makeSessionInactiveError;
+exports.getTraceSpans = getTraceSpans;
+exports.expectTracedTools = expectTracedTools;
+exports.setOryEnv = setOryEnv;
+exports.useTempConfigDir = useTempConfigDir;
+exports.clearOryEnv = clearOryEnv;
+exports.stubSessionSuccess = stubSessionSuccess;
+exports.stubSessionInactive = stubSessionInactive;
+exports.stubSessionNetworkError = stubSessionNetworkError;
+exports.stubSessionMfaRequired = stubSessionMfaRequired;
+exports.stubSessionExpired = stubSessionExpired;
+exports.stubOAuth2Success = stubOAuth2Success;
+exports.stubOAuth2Inactive = stubOAuth2Inactive;
+exports.stubPermissionAllowed = stubPermissionAllowed;
+exports.stubPermissionDenied = stubPermissionDenied;
+exports.stubPermissionNetworkError = stubPermissionNetworkError;
+exports.stubPermissionRateLimited = stubPermissionRateLimited;
+exports.spyRelationshipCreated = spyRelationshipCreated;
+exports.spyRelationshipCreateError = spyRelationshipCreateError;
+exports.stubMcpAllowed = stubMcpAllowed;
+exports.stubMcpServerDenied = stubMcpServerDenied;
+exports.stubMcpToolDenied = stubMcpToolDenied;
+exports.stubMcpNetworkError = stubMcpNetworkError;
+const fs = __importStar(require("node:fs"));
+const os = __importStar(require("node:os"));
+const path = __importStar(require("node:path"));
+const vitest_1 = require("vitest");
+const client_js_1 = require("./client.js");
+// Shared per-harness contract scenarios (session fail-open, gate
+// translation, audit-only). See contract-suite.ts for the adapter shape.
+var contract_suite_js_1 = require("./contract-suite.js");
+Object.defineProperty(exports, "runHarnessContractSuite", { enumerable: true, get: function () { return contract_suite_js_1.runHarnessContractSuite; } });
+// ─── Client Helpers ────────────────────────────────────────────────
+/**
+ * Create an OryAgentClient with session caching disabled.
+ * Pass overrides to customize (e.g. a different harness name).
+ */
+function createMockClient(overrides) {
+    return new client_js_1.OryAgentClient({
+        projectUrl: "https://test.projects.oryapis.com",
+        apiKey: "test-api-key",
+        harness: "test",
+        sessionCacheTtlMs: 0,
+        ...overrides,
+    });
+}
+/**
+ * Stub an internal API instance method on the client.
+ * Returns a vi.fn mock so callers can assert on calls.
+ */
+function stubApi(client, api, method, impl) {
+    const instance = client[api];
+    instance[method] = vitest_1.vi.fn(impl);
+    return instance[method];
+}
+// ─── Mock API Responses ────────────────────────────────────────────
+/** Successful session verification response (wraps Ory API shape). */
+exports.MOCK_SESSION_RESPONSE = {
+    data: {
+        id: "session-abc",
+        active: true,
+        authenticated_at: "2026-04-17T10:00:00Z",
+        expires_at: "2026-04-18T10:00:00Z",
+        authenticator_assurance_level: "aal1",
+        authentication_methods: [
+            { method: "password", completed_at: "2026-04-17T10:00:00Z" },
+        ],
+        identity: {
+            id: "identity-123",
+            traits: { email: "dev@example.com" },
+        },
+    },
+};
+/** Inactive session (same shape, active: false). */
+exports.MOCK_INACTIVE_SESSION_RESPONSE = {
+    data: { ...exports.MOCK_SESSION_RESPONSE.data, active: false },
+};
+/** Successful OAuth2 token introspection response. */
+exports.MOCK_OAUTH2_RESPONSE = {
+    data: {
+        active: true,
+        client_id: "agent-client",
+        sub: "user:alice",
+        scope: "tools:invoke",
+        aud: ["api"],
+        exp: 1700000000,
+        iat: 1699999000,
+    },
+};
+/** Inactive OAuth2 token response. */
+exports.MOCK_INACTIVE_OAUTH2_RESPONSE = {
+    data: { active: false },
+};
+/** Permission check allowed response. */
+exports.PERMISSION_ALLOWED = { data: { allowed: true } };
+/** Permission check denied response. */
+exports.PERMISSION_DENIED = { data: { allowed: false } };
+/** Batch permission check: both results allowed. */
+exports.BATCH_BOTH_ALLOWED = {
+    data: { results: [{ allowed: true }, { allowed: true }] },
+};
+/** Batch permission check: server allowed, tool denied. */
+exports.BATCH_SERVER_ALLOWED_TOOL_DENIED = {
+    data: { results: [{ allowed: true }, { allowed: false }] },
+};
+/** Batch permission check: server denied. */
+exports.BATCH_SERVER_DENIED = {
+    data: { results: [{ allowed: false }, { allowed: true }] },
+};
+// ─── Error Factories ───────────────────────────────────────────────
+/** Create an Axios-shaped error with a response. */
+function makeAxiosError(status, body, code) {
+    const err = {
+        isAxiosError: true,
+        message: `Request failed with status code ${status}`,
+        response: { status, data: body ?? {} },
+    };
+    if (code)
+        err.code = code;
+    return err;
+}
+/** Create a Node.js network error (ECONNREFUSED, ETIMEDOUT, etc.). */
+function makeNetworkError(code = "ECONNREFUSED") {
+    const err = new Error(`connect ${code}`);
+    err.code = code;
+    return err;
+}
+/** Axios 429 rate-limit error. */
+function makeRateLimitError() {
+    return makeAxiosError(429);
+}
+/** Axios 403 with session_aal2_required error id. */
+function makeMfaRequiredError() {
+    return makeAxiosError(403, {
+        error: { id: "session_aal2_required", message: "MFA required" },
+    });
+}
+/** Axios 401 with session_inactive error id. */
+function makeSessionInactiveError() {
+    return makeAxiosError(401, {
+        error: { id: "session_inactive", message: "No active session" },
+    });
+}
+// ─── Trace Span Assertions ────────────────────────────────────────
+/**
+ * Get all recorded trace spans from a client, optionally filtered by event.
+ */
+function getTraceSpans(client, event) {
+    const all = client.tracer.spans();
+    return event ? all.filter((s) => s.event === event) : all;
+}
+/**
+ * Assert the tracer recorded the expected tool names for a given event type.
+ * Checks the `toolName` attribute on each span.
+ */
+function expectTracedTools(client, event, expectedTools) {
+    const spans = getTraceSpans(client, event);
+    const tools = spans.map((s) => s.attributes?.toolName);
+    for (const tool of expectedTools) {
+        if (!tools.includes(tool)) {
+            throw new Error(`Expected trace event "${event}" for tool "${tool}" but got: [${tools.join(", ")}]`);
+        }
+    }
+    if (spans.length !== expectedTools.length) {
+        throw new Error(`Expected ${expectedTools.length} "${event}" spans but got ${spans.length}: [${tools.join(", ")}]`);
+    }
+}
+// ─── Environment Helpers ───────────────────────────────────────────
+/**
+ * Set standard Ory env vars for a configured + authenticated test.
+ * Returns a cleanup function that restores the previous state.
+ */
+function setOryEnv(overrides = {}) {
+    const saved = {};
+    const vars = {
+        ORY_PROJECT_URL: "projectUrl" in overrides
+            ? overrides.projectUrl
+            : "https://test.projects.oryapis.com",
+        ORY_SESSION_TOKEN: "sessionToken" in overrides ? overrides.sessionToken : "valid-token",
+        ORY_OAUTH2_TOKEN: overrides.oauth2Token,
+        ORY_AGENT_SUBJECT_ID: overrides.subjectId,
+        ORY_PERMISSION_NAMESPACE: overrides.namespace,
+        ORY_PERMISSION_MODE: "permissionMode" in overrides ? overrides.permissionMode : "enforce",
+    };
+    for (const [key, value] of Object.entries(vars)) {
+        saved[key] = process.env[key];
+        if (value !== undefined) {
+            process.env[key] = value;
+        }
+        else {
+            delete process.env[key];
+        }
+    }
+    return () => {
+        for (const [key, value] of Object.entries(saved)) {
+            if (value !== undefined) {
+                process.env[key] = value;
+            }
+            else {
+                delete process.env[key];
+            }
+        }
+    };
+}
+/**
+ * Point `XDG_CONFIG_HOME` at a fresh temp directory so `resolveConfig()`
+ * reads from a known-empty state instead of the developer's real
+ * `~/.config/ory-agent-plugins/config.json`. Returns a cleanup function
+ * that restores the env var and removes the temp directory.
+ *
+ * Use `saveConfig(...)` from `./config.js` inside the test to shape the
+ * isolated config (e.g. `saveConfig({ auditOnly: true })`).
+ */
+function useTempConfigDir() {
+    const saved = process.env.XDG_CONFIG_HOME;
+    const tmpHome = fs.mkdtempSync(path.join(os.tmpdir(), "ory-test-config-"));
+    process.env.XDG_CONFIG_HOME = tmpHome;
+    return () => {
+        if (saved !== undefined)
+            process.env.XDG_CONFIG_HOME = saved;
+        else
+            delete process.env.XDG_CONFIG_HOME;
+        fs.rmSync(tmpHome, { recursive: true, force: true });
+    };
+}
+/**
+ * Clear all Ory env vars (simulate unconfigured state).
+ */
+function clearOryEnv() {
+    delete process.env.ORY_PROJECT_URL;
+    delete process.env.ORY_API_KEY;
+    delete process.env.ORY_SESSION_TOKEN;
+    delete process.env.ORY_OAUTH2_TOKEN;
+    delete process.env.ORY_AGENT_SUBJECT_ID;
+    delete process.env.ORY_USER_SUBJECT_ID;
+    delete process.env.ORY_USER_SUBJECT_NAMESPACE;
+    delete process.env.ORY_PERMISSION_NAMESPACE;
+    // Intentionally leave ORY_PERMISSION_MODE alone — it's typically set
+    // once at the top of a test file to pin the historical enforce-mode
+    // behavior, and unconfigured-Ory tests don't exercise permission
+    // checks anyway.
+}
+// ─── Stub Presets ──────────────────────────────────────────────────
+/** Stub verifySession to succeed. */
+function stubSessionSuccess(client) {
+    return stubApi(client, "frontend", "toSession", () => Promise.resolve(exports.MOCK_SESSION_RESPONSE));
+}
+/** Stub verifySession to return inactive. */
+function stubSessionInactive(client) {
+    return stubApi(client, "frontend", "toSession", () => Promise.resolve(exports.MOCK_INACTIVE_SESSION_RESPONSE));
+}
+/** Stub verifySession to throw a network error. */
+function stubSessionNetworkError(client) {
+    return stubApi(client, "frontend", "toSession", () => Promise.reject(makeNetworkError()));
+}
+/** Stub verifySession to throw MFA required. */
+function stubSessionMfaRequired(client) {
+    return stubApi(client, "frontend", "toSession", () => Promise.reject(makeMfaRequiredError()));
+}
+/** Stub verifySession to throw session_inactive. */
+function stubSessionExpired(client) {
+    return stubApi(client, "frontend", "toSession", () => Promise.reject(makeSessionInactiveError()));
+}
+/** Stub introspectToken to succeed. */
+function stubOAuth2Success(client) {
+    return stubApi(client, "oauth2", "introspectOAuth2Token", () => Promise.resolve(exports.MOCK_OAUTH2_RESPONSE));
+}
+/** Stub introspectToken to return inactive. */
+function stubOAuth2Inactive(client) {
+    return stubApi(client, "oauth2", "introspectOAuth2Token", () => Promise.resolve(exports.MOCK_INACTIVE_OAUTH2_RESPONSE));
+}
+/** Stub checkPermission to allow. */
+function stubPermissionAllowed(client) {
+    return stubApi(client, "permission", "checkPermission", () => Promise.resolve(exports.PERMISSION_ALLOWED));
+}
+/** Stub checkPermission to deny. */
+function stubPermissionDenied(client) {
+    return stubApi(client, "permission", "checkPermission", () => Promise.resolve(exports.PERMISSION_DENIED));
+}
+/** Stub checkPermission to throw a network error. */
+function stubPermissionNetworkError(client) {
+    return stubApi(client, "permission", "checkPermission", () => Promise.reject(makeNetworkError()));
+}
+/** Stub checkPermission to throw a rate-limit error. */
+function stubPermissionRateLimited(client) {
+    return stubApi(client, "permission", "checkPermission", () => Promise.reject(makeRateLimitError()));
+}
+/**
+ * Spy on the public `createRelationship` method and resolve successfully.
+ * Prefer this over stubbing the internal relationship API: `setAgentPrincipal`
+ * rebuilds the API instances when the agent token changes (e.g. inside
+ * `sessionStart`), which silently discards API-level stubs.
+ */
+function spyRelationshipCreated(client) {
+    return vitest_1.vi
+        .spyOn(client, "createRelationship")
+        .mockResolvedValue({ created: true, alreadyExisted: false });
+}
+/** Spy on the public `createRelationship` method and reject. */
+function spyRelationshipCreateError(client) {
+    return vitest_1.vi
+        .spyOn(client, "createRelationship")
+        .mockRejectedValue(new Error("relationship write failed"));
+}
+// ─── MCP Stub Presets ─────────────────────────────────────────────
+/** Stub both checkPermission (server allow) and batchCheckPermission (both allow). */
+function stubMcpAllowed(client) {
+    stubApi(client, "permission", "checkPermission", () => Promise.resolve(exports.PERMISSION_ALLOWED));
+    return stubApi(client, "permission", "batchCheckPermission", () => Promise.resolve(exports.BATCH_BOTH_ALLOWED));
+}
+/** Stub checkPermission to deny (server-only MCP check). */
+function stubMcpServerDenied(client) {
+    return stubApi(client, "permission", "checkPermission", () => Promise.resolve(exports.PERMISSION_DENIED));
+}
+/** Stub batchCheckPermission: server allowed, tool denied. */
+function stubMcpToolDenied(client) {
+    stubApi(client, "permission", "checkPermission", () => Promise.resolve(exports.PERMISSION_ALLOWED));
+    return stubApi(client, "permission", "batchCheckPermission", () => Promise.resolve(exports.BATCH_SERVER_ALLOWED_TOOL_DENIED));
+}
+/** Stub checkPermission to throw network error (MCP fail-open). */
+function stubMcpNetworkError(client) {
+    return stubApi(client, "permission", "checkPermission", () => Promise.reject(makeNetworkError()));
+}

package/dist/tool-catalog.d.ts

@@ -16,10 +16,17 @@
  */
 export declare const HARNESS_TOOL_CATALOG: {
     readonly "claude-code": readonly ["Read", "Write", "Edit", "Bash", "Glob", "Grep", "WebFetch", "WebSearch", "NotebookEdit", "TodoWrite", "Task"];
+    readonly "claude-agent-sdk": readonly ["Read", "Write", "Edit", "Bash", "Glob", "Grep", "WebFetch", "WebSearch", "NotebookEdit", "TodoWrite", "Task"];
     readonly codex: readonly ["shell", "apply_patch", "read_file"];
     readonly "gemini-cli": readonly ["read_file", "write_file", "edit_file", "list_files", "search_files", "shell", "web_search"];
     readonly openclaw: readonly ["execute_command", "read_file", "write_file", "list_directory", "search_files", "browser"];
     readonly opencode: readonly ["read", "write", "edit", "bash", "glob", "grep", "webfetch"];
+    readonly continue: readonly ["Bash", "Read", "Edit", "Write", "Grep", "Glob", "WebSearch"];
+    readonly goose: readonly ["shell", "text_editor", "read_file", "write_file", "list_windows", "screen_capture"];
+    readonly cline: readonly ["execute_command", "read_file", "write_to_file", "replace_in_file", "search_files", "list_files", "use_mcp_tool"];
+    readonly amp: readonly ["Bash", "Read", "create_file", "edit_file", "undo_edit", "glob", "Grep", "finder", "read_web_page", "web_search", "todo_read", "todo_write", "oracle", "Task"];
+    readonly pi: readonly ["read", "write", "edit", "bash", "grep", "find", "ls"];
+    readonly antigravity: readonly ["run_command", "view_file", "write_to_file", "replace_file_content", "multi_replace_file_content", "list_directory", "grep_search", "find", "read_url_content", "search_web", "call_mcp_tool"];
 };
 /** Names of the harnesses with a known tool catalog. */
 export type KnownHarness = keyof typeof HARNESS_TOOL_CATALOG;

package/dist/tool-catalog.js

@@ -34,6 +34,22 @@ exports.HARNESS_TOOL_CATALOG = {
         "TodoWrite",
         "Task",
     ],
+    // The Claude Agent SDK exposes the same built-in tool surface as Claude
+    // Code (it is the embeddable form of the same runtime), so the catalog
+    // mirrors claude-code's and `permissions bootstrap` can seed it.
+    "claude-agent-sdk": [
+        "Read",
+        "Write",
+        "Edit",
+        "Bash",
+        "Glob",
+        "Grep",
+        "WebFetch",
+        "WebSearch",
+        "NotebookEdit",
+        "TodoWrite",
+        "Task",
+    ],
     codex: ["shell", "apply_patch", "read_file"],
     "gemini-cli": [
         "read_file",
@@ -53,6 +69,54 @@ exports.HARNESS_TOOL_CATALOG = {
         "browser",
     ],
     opencode: ["read", "write", "edit", "bash", "glob", "grep", "webfetch"],
+    continue: ["Bash", "Read", "Edit", "Write", "Grep", "Glob", "WebSearch"],
+    goose: [
+        "shell",
+        "text_editor",
+        "read_file",
+        "write_file",
+        "list_windows",
+        "screen_capture",
+    ],
+    cline: [
+        "execute_command",
+        "read_file",
+        "write_to_file",
+        "replace_in_file",
+        "search_files",
+        "list_files",
+        "use_mcp_tool",
+    ],
+    amp: [
+        "Bash",
+        "Read",
+        "create_file",
+        "edit_file",
+        "undo_edit",
+        "glob",
+        "Grep",
+        "finder",
+        "read_web_page",
+        "web_search",
+        "todo_read",
+        "todo_write",
+        "oracle",
+        "Task",
+    ],
+    pi: ["read", "write", "edit", "bash", "grep", "find", "ls"],
+    antigravity: [
+        "run_command",
+        "view_file",
+        "write_to_file",
+        "replace_file_content",
+        "multi_replace_file_content",
+        "list_directory",
+        "grep_search",
+        "find",
+        "read_url_content",
+        "search_web",
+        "call_mcp_tool",
+    ],
 };
 exports.KNOWN_HARNESSES = Object.keys(exports.HARNESS_TOOL_CATALOG);
 /**
@@ -88,10 +152,19 @@ exports.ALL_TOOLS = Array.from(new Set(Object.values(exports.HARNESS_TOOL_CATALO
  */
 exports.INTERACTIVE_TOOL_CATALOG = {
     "claude-code": ["AskUserQuestion", "ExitPlanMode", "TodoWrite"],
+    // Same runtime as Claude Code — the SDK delivers the same
+    // user-interaction primitives through PreToolUse.
+    "claude-agent-sdk": ["AskUserQuestion", "ExitPlanMode", "TodoWrite"],
     codex: [],
     "gemini-cli": [],
     openclaw: [],
     opencode: [],
+    continue: [],
+    goose: [],
+    cline: [],
+    amp: [],
+    pi: [],
+    antigravity: [],
 };
 function parseEnvInteractiveTools() {
     const raw = process.env.ORY_INTERACTIVE_TOOLS;

package/dist/user-login.js

@@ -165,8 +165,13 @@ async function runUserLogin(client, options) {
             };
         }
     }
-    // 2. Pre-supplied env tokens short-circuit the browser flow (CI, scripted runs).
-    if (process.env.ORY_SESSION_TOKEN || process.env.ORY_OAUTH2_TOKEN) {
+    // 2. Pre-supplied env tokens short-circuit the browser flow (CI, scripted
+    //    runs). ORY_USER_* are the documented names; the legacy single-identity
+    //    ORY_SESSION_TOKEN / ORY_OAUTH2_TOKEN are still honored.
+    if (process.env.ORY_USER_SESSION_TOKEN ||
+        process.env.ORY_USER_OAUTH2_TOKEN ||
+        process.env.ORY_SESSION_TOKEN ||
+        process.env.ORY_OAUTH2_TOKEN) {
         return {
             proceed: true,
             mode: "env_token",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ory/argus",
-  "version": "0.9.1",
+  "version": "0.11.0",
   "description": "Ory Argus: the core API for building authentication, authorization, and audit into AI agent harness plugins, extensions, and custom integrations",
   "license": "Apache-2.0",
   "homepage": "https://ory.com",