@beryl-so/cli 0.41.0 → 0.42.0

package/dist/adapters/mcp.js

@@ -10,6 +10,7 @@ import { ApiClient, ApiError } from "../http.js";
 import { parseArgv } from "./cli.js";
 import { commands } from "../registry/index.js";
 import { analyticsOptions, createPostHogClient, setClient, setSurface, shutdownTelemetry, withTool, } from "../telemetry.js";
+import { refreshInstalledSkills } from "../commands/init.js";
 import { cliVersion, warnIfStale } from "../version-check.js";
 export function toolName(spec) {
     return spec.name.replace(/ /g, "_").replace(/-/g, "_");
@@ -292,6 +293,17 @@ export async function serveMcp(baseCtx) {
     // Fire-and-forget staleness warning: a stale MCP server silently exposes fewer
     // tools, and stderr is the one channel a stdio MCP server can safely log to.
     void warnIfStale(cliVersion(), (msg) => console.error(msg));
+    // Installed skill copies only refresh on `init`; this server is respawned from @latest
+    // every session, so bring them up to this version here. Never fatal.
+    try {
+        for (const { outcome, file } of refreshInstalledSkills(process.cwd())) {
+            if (outcome !== "unchanged")
+                console.error(`beryl-test authoring skill ${outcome}: ${file}`);
+        }
+    }
+    catch {
+        /* best-effort */
+    }
     const server = new Server({ name: "beryl", version: cliVersion() }, {
         capabilities: { tools: {} },
         instructions: mcpInstructions(),

package/dist/beryl-test-skill.js

@@ -93,10 +93,10 @@ Then, per test:
 2. **Write the ActionPlan**, with one real outcome assertion (§2).
 3. \`beryl tests lint --file plan.json\` — schema check, offline, no network.
 4. \`beryl tests create --title … --file … --description "<the intent>"\` — replays the
-   plan locally, banks it only on green; a red replay hands back the failure — fix and
-   re-run (§3 for the intent, §4 for the loop).
-5. \`beryl runs local\` — re-run banked tests on your machine while iterating.
-6. \`beryl runs trigger\` — hand it to Beryl's cloud, on demand or on a schedule.
+   plan locally, banks it only on green, records no run; a red replay hands back the
+   failure — fix and re-run (§3 for the intent, §4 for the loop).
+5. When the batch is banked, run the new ids together ONCE (\`beryl runs local <id…>\`,
+   or \`beryl runs trigger\` for the cloud) so they land as one run with real first results.
 
 ## 1. First: who do your tests sign in as?
 
@@ -285,8 +285,8 @@ ${JSON.stringify(BERYL_TEST_SKILL_EXAMPLE_PLAN, null, 2)
      --description "<the intent — see §3>"
    \`\`\`
    \`create\` banks the plan only on a green local replay; a red one hands back the
-   failure — fix and re-run. Replay mechanics (where it runs over MCP, \`--no-sync\`,
-   \`--no-verify\`, the cloud-session exception) are in \`tests create\`'s own
+   failure — fix and re-run. It records no run. Replay mechanics (where it runs over
+   MCP, \`--no-verify\`, the cloud-session exception) are in \`tests create\`'s own
    description. Inspect the exact spec with \`beryl tests script --file plan.json\`.
    Full ActionPlan JSON Schema:
    https://api.beryl.so/api/v1/schemas/action-plan.schema.json.
@@ -378,14 +378,12 @@ beryl runs local <test-id> --no-sync --url-override http://localhost:3000 --dir
 beryl runs local              # the whole suite, results recorded in Beryl
 \`\`\`
 
-The loop: draft → \`tests lint\` → \`tests create\` → \`runs local --no-sync\` while
-iterating (a synced run lands in the project's run history) → read \`--dir\`'s
-\`report.json\` to see which step or assertion failed and why → fix the plan →
-\`tests set-plan\` → run again. Flag semantics, session-mode behavior, \`await_email\`
-and authenticated-test handling, and exit codes are in \`runs local\`'s own description.
-
-Once the test passes locally against a real outcome, it's ready to bank and let Beryl
-run and heal it.
+The loop: draft → \`tests lint\` → \`tests create\` (proves, records nothing) →
+\`runs local --no-sync\` while iterating → read \`--dir\`'s \`report.json\` to see which
+step or assertion failed and why → fix the plan → \`tests set-plan\` → run again. When
+the batch is done, one synced \`runs local <id…>\` (or \`runs trigger\`) records it as a
+single run. Flag semantics, session-mode behavior, \`await_email\` and
+authenticated-test handling, and exit codes are in \`runs local\`'s own description.
 
 ## 5. Testing an OTP / signup flow (\`await_email\`)
 

package/dist/commands/init.js

@@ -118,6 +118,16 @@ function writeSkills(cwd, scope) {
         writeSkillFile(skillLeaf(path.join(root, ".claude", "skills"))),
     ];
 }
+// The MCP server is respawned from `npx @beryl-so/cli@latest` at every session start, so
+// its startup is the one moment new CLI code reliably runs on a user's machine. Refresh
+// whatever copies `init` installed; never create one (scope is `init`'s call).
+export function refreshInstalledSkills(cwd) {
+    const files = new Set([os.homedir(), cwd].flatMap((root) => [
+        skillLeaf(path.join(root, ".agents", "skills")),
+        skillLeaf(path.join(root, ".claude", "skills")),
+    ]));
+    return [...files].filter((file) => fs.existsSync(file)).map(writeSkillFile);
+}
 // Local authoring drives a real browser, and the whole authoring workflow (walk the flow
 // first, then bank the plan) depends on it — so the install is mandatory, not offered:
 // missing means install now, no prompt, no opt-out. The runner ships with the CLI; the

package/dist/commands/tests.js

@@ -2,7 +2,7 @@ import fs from "node:fs";
 import { CliError, UsageError } from "../errors.js";
 import { ApiError } from "../http.js";
 import { lintPlan } from "../lint.js";
-import { buildImportForm, establishAccountSession, executeLocalSpec, toRunEntry, } from "../local-exec.js";
+import { establishAccountSession, executeLocalSpec, } from "../local-exec.js";
 import { PlaywrightMissingError } from "../local-run.js";
 import { dim, green, red, table, yellow } from "../output.js";
 import { confirmInstall, installCommandsFor, installPlaywright, playwrightGaps, } from "../playwright-install.js";
@@ -216,8 +216,9 @@ export const testCommands = [
             "Beryl's; if the Chromium browser is missing there the tool returns the install command " +
             "(`beryl init` installs it; on a terminal the CLI offers to). " +
             "A red replay banks NOTHING: the failure evidence comes back (over MCP the screenshot is " +
-            "image content), you fix the plan file and re-run. The proving run is imported as the " +
-            "test's first run (--no-sync to skip). A plan that signs in with a session Beryl captured " +
+            "image content), you fix the plan file and re-run. No run is recorded: once a batch of " +
+            "tests is banked, run their ids together once (`runs local`, or `runs trigger` for the " +
+            "cloud) so they land as ONE run with real first results. A plan that signs in with a session Beryl captured " +
             "server-side cannot replay locally (that session never leaves Beryl's cloud). It falls " +
             "back to server-side verification automatically, and says so. A session-mode plan replays locally " +
             "fine: the server renders it with its account's stored sign-in steps in front, so " +
@@ -258,13 +259,6 @@ export const testCommands = [
                     "name is an error (create groups with `beryl groups create`). Omit for no group.",
             },
             { name: "env", type: "string", description: "Environment id to compile and prove against" },
-            {
-                name: "sync",
-                type: "boolean",
-                default: true,
-                description: "Import the green proving replay as the test's first run (--no-sync: bank only, " +
-                    "no run recorded)",
-            },
             {
                 name: "dir",
                 type: "string",
@@ -285,7 +279,6 @@ export const testCommands = [
             const description = flagStr(input, "description");
             const urlOverride = flagStr(input, "url-override");
             const env = flagStr(input, "env");
-            const sync = input.flags.sync !== false;
             const groups = flagStrings(input, "group");
             const bank = (extra) => ctx.client.post(`${projectPath(workspaceId, projectId)}/tests`, {
                 title,
@@ -306,9 +299,6 @@ export const testCommands = [
                 title,
                 base_url: urlOverride,
                 environment_id: env,
-                // The proof replay becomes the test's first run, so capture the filmstrip
-                // frames its replay view needs.
-                frames: sync ? true : undefined,
             });
             let compiled = await compile();
             // Session-mode plan: establish the account's session once (server-side, reused
@@ -370,7 +360,6 @@ export const testCommands = [
                 inlinedLoginSteps: compiled.inlined_login_steps ?? 0,
                 email: compiled.email ?? undefined,
             });
-            const startedAt = new Date().toISOString();
             const attempts = 1 + Math.max(0, compiled.policy.error_retries);
             let outcome;
             let runError;
@@ -457,31 +446,13 @@ export const testCommands = [
                     return conflict;
                 throw err;
             }
-            let runId;
-            if (sync) {
-                const entry = toRunEntry({ ...specOf(compiled.content), id: String(created.id) }, outcome, undefined, startedAt, 0, [...(loginPassword ? [loginPassword] : []), ...(session?.secrets ?? [])]);
-                const imported = (await ctx.client.request("POST", `${projectPath(workspaceId, projectId)}/runs/import`, {
-                    form: buildImportForm([entry], {
-                        environmentId: env,
-                        targetUrlOverride: urlOverride,
-                        // The proof replay is authoring machinery, not a run the customer
-                        // asked for — it must never email/Slack "your report is ready".
-                        notifications: false,
-                        startedAt,
-                        completedAt: new Date().toISOString(),
-                        onNote: (line) => ctx.err(yellow(`! ${line}`)),
-                    }),
-                }));
-                runId = imported.id;
-            }
             if (urlOverride) {
                 ctx.err(yellow(`! The proof ran against ${urlOverride}, not the environment's root URL — ` +
                     "the banked test is unproven against its real target until a run there passes."));
             }
             return {
-                data: { ...created, ...(runId ? { proof_run_id: runId } : {}) },
-                human: green(`✓ Replay passed locally — banked as ${created.id}`) +
-                    (runId ? `\nProving run imported: \`beryl runs get ${runId}\`` : ""),
+                data: created,
+                human: green(`✓ Replay passed locally — banked as ${created.id}`),
             };
         },
     },

package/dist/local-exec.js

@@ -117,7 +117,6 @@ export function buildImportForm(entries, opts) {
     const manifest = {
         environment_id: opts.environmentId ?? null,
         target_url_override: opts.targetUrlOverride ?? null,
-        ...(opts.notifications !== undefined ? { notifications: opts.notifications } : {}),
         started_at: opts.startedAt,
         completed_at: opts.completedAt,
         results: entries.map(({ files: _files, ...rest }) => rest),

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@beryl-so/cli",
-  "version": "0.41.0",
+  "version": "0.42.0",
   "description": "Beryl on the command line — projects, runs, the exploring agent, and an MCP server over the same commands.",
   "license": "MIT",
   "type": "module",