@beryl-so/cli 0.5.0 → 0.6.0

package/README.md

@@ -73,6 +73,14 @@ Authenticate the CLI with your Beryl account
 | --- | --- | --- |
 | `beryl login` | Authenticate the CLI with your Beryl account | — |
 
+### signup
+
+Register a new Beryl account (emails a 6-digit verification code)
+
+| Command | Summary | MCP tool |
+| --- | --- | --- |
+| `beryl signup` | Register a new Beryl account (emails a 6-digit verification code) | `signup` |
+
 ### logout
 
 Remove the stored token from the CLI config
@@ -232,6 +240,7 @@ Inspect the agent's exploration runs — how it crawled a site and authored its
 | `beryl explorations list` | List the agent's exploration passes for a project | `explorations_list` |
 | `beryl explorations get <exploration-id>` | Show one exploration: authored tests, abandoned flows, coverage, frontier | `explorations_get` |
 | `beryl explorations steps <exploration-id>` | List every step the agent took in an exploration | `explorations_steps` |
+| `beryl explorations cancel <exploration-id>` | Cancel an in-flight exploration | `explorations_cancel` |
 | `beryl explorations watch <exploration-id>` | Stream an exploration live — watch the agent explore and author tests | `explorations_watch` |
 
 ### config
@@ -251,6 +260,19 @@ Manage the variables, secrets, and files the agent can use while exploring and r
 | `beryl config files download <file-id>` | Get a short-lived download URL for a file | `config_files_download` |
 | `beryl config files delete <file-id>` | Delete an uploaded file | `config_files_delete` |
 
+### slack
+
+Send run outcomes to a Slack channel via a per-project incoming webhook.
+
+`beryl slack` with no subcommand runs `slack show`.
+
+| Command | Summary | MCP tool |
+| --- | --- | --- |
+| `beryl slack show` | Show the project's Slack alert config (webhook is masked) | `slack_show` |
+| `beryl slack set-webhook <url>` | Set the Slack incoming-webhook URL run alerts post to | `slack_set_webhook` |
+| `beryl slack clear` | Remove the project's Slack webhook (stops all alerts) | `slack_clear` |
+| `beryl slack test` | Post a sample alert to the configured webhook | `slack_test` |
+
 ### credentials
 
 Manage saved logins Beryl reuses to test behind authentication, and attach them to projects.
@@ -273,10 +295,21 @@ Drive a browser session that captures a target-site login for Beryl to reuse.
 | Command | Summary | MCP tool |
 | --- | --- | --- |
 | `beryl auth-capture start` | Start a login-capture browser session for the project (non-interactive) | `auth_capture_start` |
+| `beryl auth-capture login <session-id>` | Log into the target site headlessly with credentials (no human at the browser) | `auth_capture_login` |
 | `beryl auth-capture capture <session-id>` | Capture the session after the user has logged in via the live-view URL | `auth_capture_capture` |
 | `beryl auth-capture refresh <session-id>` | Capture a refreshed session for a project whose login is expiring | `auth_capture_refresh` |
 | `beryl auth-capture release <session-id>` | Release a login-capture browser session without capturing | `auth_capture_release` |
 
+### inbox
+
+Disposable email inboxes for testing flows that send mail — signups, OTPs, receipts.
+
+| Command | Summary | MCP tool |
+| --- | --- | --- |
+| `beryl inbox create` | Mint a disposable email inbox that Beryl receives mail for | `inbox_create` |
+| `beryl inbox read <inbox-id>` | Read the latest email from an inbox (waits for one to arrive) | `inbox_read` |
+| `beryl inbox emails <inbox-id>` | List the emails an inbox has received | `inbox_emails` |
+
 ### account
 
 View and update your personal account profile.

package/dist/commands/auth.js

@@ -54,7 +54,9 @@ export const authCommands = [
         summary: "Authenticate the CLI with your Beryl account",
         description: "Signs in with an emailed one-time code and mints a personal access token, " +
             "which is stored in the CLI config. Pass --token to use an existing token " +
-            "from Account → API tokens instead. In CI, prefer the BERYL_API_KEY environment variable.",
+            "from Account → API tokens instead. Pass --email plus --code (the 6 digits " +
+            "from the email, e.g. read from a `beryl inbox`) to complete the OTP flow " +
+            "without a prompt. In CI, prefer the BERYL_API_KEY environment variable.",
         interactive: true,
         flags: [
             { name: "token", type: "string", description: "Use an existing personal access token" },
@@ -64,26 +66,48 @@ export const authCommands = [
                 description: "Email for the one-time code sign-in (default: your last sign-in, or your git " +
                     "user.email — offered as the prompt default)",
             },
+            {
+                name: "code",
+                type: "string",
+                description: "The emailed 6-digit code — skips the prompt for non-interactive use " +
+                    "(requires --email; also skips sending a fresh code, so pair it with a " +
+                    "code already requested via `beryl login`, `beryl signup`, or the API)",
+            },
             {
                 name: "token-name",
                 type: "string",
                 description: "Name for the minted token (default: CLI on <hostname>)",
             },
         ],
-        examples: ["beryl login", "beryl login --token beryl_pat_…", "beryl login --email you@example.com"],
+        examples: [
+            "beryl login",
+            "beryl login --token beryl_pat_…",
+            "beryl login --email you@example.com",
+            "beryl login --email agent@example.com --code 123456 --json",
+        ],
         async run(ctx, input) {
             const apiUrl = ctx.client.baseUrl;
             let token = flagStr(input, "token");
             if (!token) {
+                const codeFlag = flagStr(input, "code")?.replace(/\s/g, "");
+                if (codeFlag !== undefined) {
+                    if (!/^\d{6}$/.test(codeFlag))
+                        throw new UsageError("--code must be the 6 digits from the email");
+                    if (!flagStr(input, "email"))
+                        throw new UsageError("--code requires --email");
+                }
                 const email = flagStr(input, "email") ?? (await promptForEmail(ctx));
                 if (!email.includes("@"))
                     throw new UsageError(`"${email}" is not an email address`);
                 const anon = new ApiClient(apiUrl);
-                await anon.post("/auth/request-login-otp", { email });
-                ctx.err(dim(`Sent a 6-digit code to ${email}`));
-                ctx.err(dim("Not arriving? Check spam; if you're new to Beryl, sign up at https://beryl.so first — " +
-                    "or use `beryl login --token` with a token from beryl.so → Account → API tokens."));
-                const code = await promptForOtpCode(ctx);
+                let code = codeFlag;
+                if (!code) {
+                    await anon.post("/auth/request-login-otp", { email });
+                    ctx.err(dim(`Sent a 6-digit code to ${email}`));
+                    ctx.err(dim("Not arriving? Check spam; if you're new to Beryl, sign up at https://beryl.so first — " +
+                        "or use `beryl login --token` with a token from beryl.so → Account → API tokens."));
+                    code = await promptForOtpCode(ctx);
+                }
                 const login = (await anon.post("/auth/verify-otp", { email, code }));
                 if (!login.access_token)
                     throw new CliError("Login did not return an access token");
@@ -125,6 +149,36 @@ export const authCommands = [
             };
         },
     },
+    {
+        name: "signup",
+        summary: "Register a new Beryl account (emails a 6-digit verification code)",
+        description: "Creates a passwordless account for the email and sends it a 6-digit code. " +
+            "Finish with `beryl login --email <addr> --code <the 6 digits>` — that verifies " +
+            "the account, creates its workspace, and signs the CLI in. With an inbox from " +
+            "`beryl inbox create` as the address, an agent can provision a fresh account " +
+            "end-to-end with no human at a prompt.",
+        flags: [
+            {
+                name: "email",
+                type: "string",
+                required: true,
+                description: "Email address for the new account",
+            },
+        ],
+        examples: ["beryl signup --email agent+ci@example.com --json"],
+        async run(ctx, input) {
+            const email = flagStr(input, "email") ?? "";
+            if (!email.includes("@"))
+                throw new UsageError(`"${email}" is not an email address`);
+            const anon = new ApiClient(ctx.client.baseUrl);
+            const created = (await anon.post("/auth/signup", { email }));
+            return {
+                data: created,
+                human: `${green("Signup started")} for ${email}\n` +
+                    dim(`Sent a 6-digit code to ${email} — finish with: beryl login --email ${email} --code <code>`),
+            };
+        },
+    },
     {
         name: "logout",
         summary: "Remove the stored token from the CLI config",

package/dist/commands/credentials.js

@@ -1,5 +1,6 @@
+import { UsageError } from "../errors.js";
 import { dim, green, yellow } from "../output.js";
-import { arg, flagBool } from "./util.js";
+import { arg, flagBool, flagStr } from "./util.js";
 const capturePath = (ws, p) => `/auth-capture/workspaces/${ws}/projects/${p}/sessions`;
 export const credentialCommands = [
     {
@@ -110,6 +111,37 @@ export const credentialCommands = [
             return { data: await ctx.client.post(capturePath(workspaceId, projectId)) };
         },
     },
+    {
+        name: "auth-capture login",
+        summary: "Log into the target site headlessly with credentials (no human at the browser)",
+        description: "Drives the login inside the capture session started by `auth-capture start`, so " +
+            "an agent can complete start → login → capture with zero human intervention. The " +
+            "credentials are sent to the server, typed into the target site over the wire, and " +
+            "never stored, logged, or returned — the captured session stays encrypted " +
+            "server-side. Follow with `auth-capture capture` to snapshot the authenticated session.",
+        scope: "project",
+        args: [
+            { name: "session-id", description: "Session id from auth-capture start", required: true },
+        ],
+        flags: [
+            { name: "username", type: "string", description: "Login username / email", required: true },
+            { name: "password", type: "string", description: "Login password", required: true },
+            {
+                name: "login-url",
+                type: "string",
+                description: "Explicit login page URL (defaults to the session's current page)",
+            },
+        ],
+        async run(ctx, input) {
+            const { workspaceId, projectId } = await ctx.requireProject(input);
+            const username = flagStr(input, "username");
+            const password = flagStr(input, "password");
+            if (!username || !password)
+                throw new UsageError("--username and --password are required");
+            await ctx.client.post(`${capturePath(workspaceId, projectId)}/${arg(input, "session-id")}/login`, { username, password, login_url: flagStr(input, "login-url") ?? null });
+            return { human: "Logged in." };
+        },
+    },
     {
         name: "auth-capture capture",
         summary: "Capture the session after the user has logged in via the live-view URL",

package/dist/commands/explorations.js

@@ -35,6 +35,20 @@ export const explorationCommands = [
             };
         },
     },
+    {
+        name: "explorations cancel",
+        summary: "Cancel an in-flight exploration",
+        description: "Stops a running exploration without touching the project or its tests — the " +
+            "lever for a runaway/non-convergent pass burning agent budget.",
+        scope: "project",
+        args: [{ name: "exploration-id", description: "Exploration id", required: true }],
+        async run(ctx, input) {
+            const { workspaceId, projectId } = await ctx.requireProject(input);
+            return {
+                data: await ctx.client.post(`${projectPath(workspaceId, projectId)}/explorations/${arg(input, "exploration-id")}/cancel`),
+            };
+        },
+    },
     {
         name: "explorations watch",
         summary: "Stream an exploration live — watch the agent explore and author tests",

package/dist/commands/inboxes.js

@@ -0,0 +1,117 @@
+import { dim, green } from "../output.js";
+import { arg, flagBool, flagNum, flagStr } from "./util.js";
+// Mirrors the API's own extractor (email_inbox/step_resolver.py): a labelled digit run
+// ("your code is 654321") beats the bare fenced pattern, because a real sign-in mail is
+// full of innocent 4-8 digit runs — "© 2026", a support number — and the bare fence
+// would happily return the first of them.
+const CODE_PATTERN = /(?<!\d)(\d{4,8})(?!\d)/;
+const LABELLED_CODE_PATTERN = /(?:verification|security|one[\s-]?time|login|sign[\s-]?in|access|confirmation)?\s*(?:code|otp|passcode|pin)\b[^0-9]{0,20}(?<!\d)(\d{4,8})(?!\d)/i;
+function visibleText(html) {
+    return html
+        .replace(/<(style|script|head)\b[\s\S]*?<\/\1>/gi, " ")
+        .replace(/<[^>]+>/g, " ");
+}
+function extractCode(email) {
+    const body = email.body_text || (email.body_html ? visibleText(email.body_html) : "");
+    for (const pattern of [LABELLED_CODE_PATTERN, CODE_PATTERN]) {
+        for (const text of [body, email.subject ?? ""]) {
+            const match = text.match(pattern);
+            if (match)
+                return match[1];
+        }
+    }
+    return null;
+}
+export const inboxCommands = [
+    {
+        name: "inbox create",
+        summary: "Mint a disposable email inbox that Beryl receives mail for",
+        groupSummary: "Disposable email inboxes for testing flows that send mail — signups, OTPs, receipts.",
+        description: "Creates a receiving address under Beryl's inbound email domain and returns it. " +
+            "Use it wherever a test needs a real, readable mailbox — e.g. as the --email for " +
+            "`beryl signup`, then read the code back with `beryl inbox read --extract-code`.",
+        scope: "workspace",
+        flags: [
+            {
+                name: "ttl-hours",
+                type: "number",
+                description: "Hours before the inbox expires and stops receiving (1-168, default 24)",
+            },
+            { name: "project", type: "string", description: "Attach the inbox to a project id" },
+        ],
+        examples: ["beryl inbox create --json", "beryl inbox create --ttl-hours 2"],
+        async run(ctx, input) {
+            const ws = await ctx.requireWorkspace(input);
+            const inbox = (await ctx.client.post(`/workspaces/${ws}/inboxes`, {
+                ttl_hours: flagNum(input, "ttl-hours") ?? 24,
+                project_id: flagStr(input, "project") ?? null,
+            }));
+            return {
+                data: inbox,
+                human: `${green("Created")} inbox ${inbox.id}\n\n  ${inbox.address}\n\n` +
+                    dim(`Read it with: beryl inbox read ${inbox.id}`),
+            };
+        },
+    },
+    {
+        name: "inbox read",
+        summary: "Read the latest email from an inbox (waits for one to arrive)",
+        description: "Waits up to --timeout-s for a matching email and returns it (one blocking request; " +
+            "the server caps the wait at 50s — re-run to keep waiting). With --extract-code, " +
+            "also pulls the one-time code (4-8 digits) out of the body/subject — handy for " +
+            "completing `beryl login --email <addr> --code <code>` unattended. Exits non-zero " +
+            "if nothing arrives before the timeout.",
+        scope: "workspace",
+        args: [{ name: "inbox-id", description: "Inbox id from `beryl inbox create`", required: true }],
+        flags: [
+            {
+                name: "timeout-s",
+                type: "number",
+                description: "Seconds to wait for a matching email (0 = don't wait; max 50, default 30)",
+            },
+            { name: "since", type: "string", description: "Only emails received after this ISO timestamp" },
+            { name: "from-contains", type: "string", description: "Only emails whose sender contains this" },
+            {
+                name: "subject-contains",
+                type: "string",
+                description: "Only emails whose subject contains this",
+            },
+            {
+                name: "extract-code",
+                type: "boolean",
+                description: "Also return the one-time code found in the email as `code`",
+            },
+        ],
+        examples: [
+            "beryl inbox read ibx_123 --timeout-s 45 --json",
+            "beryl inbox read ibx_123 --subject-contains code --extract-code --json",
+        ],
+        async run(ctx, input) {
+            const ws = await ctx.requireWorkspace(input);
+            const email = (await ctx.client.get(`/workspaces/${ws}/inboxes/${arg(input, "inbox-id")}/emails/latest`, {
+                timeout_s: flagNum(input, "timeout-s"),
+                since: flagStr(input, "since"),
+                from_contains: flagStr(input, "from-contains"),
+                subject_contains: flagStr(input, "subject-contains"),
+            }));
+            if (!flagBool(input, "extract-code"))
+                return { data: email };
+            return { data: { ...email, code: extractCode(email) } };
+        },
+    },
+    {
+        name: "inbox emails",
+        summary: "List the emails an inbox has received",
+        scope: "workspace",
+        args: [{ name: "inbox-id", description: "Inbox id from `beryl inbox create`", required: true }],
+        flags: [
+            { name: "since", type: "string", description: "Only emails received after this ISO timestamp" },
+        ],
+        async run(ctx, input) {
+            const ws = await ctx.requireWorkspace(input);
+            return {
+                data: await ctx.client.get(`/workspaces/${ws}/inboxes/${arg(input, "inbox-id")}/emails`, { since: flagStr(input, "since") }),
+            };
+        },
+    },
+];

package/dist/commands/slack.js

@@ -0,0 +1,82 @@
+import { arg, flagBool, projectPath } from "./util.js";
+const slackPath = (ws, p) => `${projectPath(ws, p)}/slack`;
+export const slackCommands = [
+    {
+        name: "slack show",
+        summary: "Show the project's Slack alert config (webhook is masked)",
+        groupSummary: "Send run outcomes to a Slack channel via a per-project incoming webhook.",
+        scope: "project",
+        groupDefault: true,
+        async run(ctx, input) {
+            const { workspaceId, projectId } = await ctx.requireProject(input);
+            return { data: await ctx.client.get(slackPath(workspaceId, projectId)) };
+        },
+    },
+    {
+        name: "slack set-webhook",
+        summary: "Set the Slack incoming-webhook URL run alerts post to",
+        scope: "project",
+        args: [
+            {
+                name: "url",
+                description: "Slack incoming-webhook URL (https://hooks.slack.com/…)",
+                required: true,
+            },
+        ],
+        flags: [
+            {
+                name: "all-runs",
+                type: "boolean",
+                description: "Post on every run, not only on a new regression",
+            },
+            {
+                name: "failures-only",
+                type: "boolean",
+                description: "Post only on a new regression (the default)",
+            },
+        ],
+        async run(ctx, input) {
+            const { workspaceId, projectId } = await ctx.requireProject(input);
+            const path = slackPath(workspaceId, projectId);
+            // Preserve the stored notify preference unless the caller states one, so
+            // rotating the webhook doesn't silently reset an existing all-runs opt-in.
+            let notifyAllRuns;
+            if (flagBool(input, "all-runs")) {
+                notifyAllRuns = true;
+            }
+            else if (flagBool(input, "failures-only")) {
+                notifyAllRuns = false;
+            }
+            else {
+                const current = (await ctx.client.get(path));
+                notifyAllRuns = current?.notify_all_runs ?? false;
+            }
+            return {
+                data: await ctx.client.put(path, {
+                    webhook_url: arg(input, "url"),
+                    notify_all_runs: notifyAllRuns,
+                }),
+            };
+        },
+    },
+    {
+        name: "slack clear",
+        summary: "Remove the project's Slack webhook (stops all alerts)",
+        scope: "project",
+        async run(ctx, input) {
+            const { workspaceId, projectId } = await ctx.requireProject(input);
+            await ctx.client.del(slackPath(workspaceId, projectId));
+            return { human: "Cleared." };
+        },
+    },
+    {
+        name: "slack test",
+        summary: "Post a sample alert to the configured webhook",
+        scope: "project",
+        async run(ctx, input) {
+            const { workspaceId, projectId } = await ctx.requireProject(input);
+            await ctx.client.post(`${slackPath(workspaceId, projectId)}/test`);
+            return { human: "Sent a test message to Slack." };
+        },
+    },
+];

package/dist/registry/index.js

@@ -4,10 +4,12 @@ import { configCommands } from "../commands/config-vars.js";
 import { credentialCommands } from "../commands/credentials.js";
 import { environmentCommands } from "../commands/environments.js";
 import { explorationCommands } from "../commands/explorations.js";
+import { inboxCommands } from "../commands/inboxes.js";
 import { initCommands } from "../commands/init.js";
 import { mcpCommands } from "../commands/mcp.js";
 import { projectCommands } from "../commands/projects.js";
 import { runCommands } from "../commands/runs.js";
+import { slackCommands } from "../commands/slack.js";
 import { testCommands } from "../commands/tests.js";
 import { workspaceCommands } from "../commands/workspaces.js";
 export const WORKSPACE_FLAG = {
@@ -43,7 +45,9 @@ export const commands = [
     ...runCommands,
     ...explorationCommands,
     ...configCommands,
+    ...slackCommands,
     ...credentialCommands,
+    ...inboxCommands,
     ...accountCommands,
     ...mcpCommands,
 ].map(withScopeFlags);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@beryl-so/cli",
-  "version": "0.5.0",
+  "version": "0.6.0",
   "description": "Beryl on the command line — projects, runs, the exploring agent, and an MCP server over the same commands.",
   "license": "MIT",
   "type": "module",