@playdrop/playdrop-cli 0.12.16 → 0.12.18

package/config/client-meta.json

@@ -1,6 +1,6 @@
 {
   "version": "0.12.8",
-  "build": 9,
+  "build": 11,
   "runtimeSdkVersion": "0.12.5",
   "runtimeSdkBuild": 2,
   "clients": {

package/dist/commands/worker/runtime.js

@@ -516,9 +516,14 @@ function buildClaudeExecArgs(input) {
         : input.playwrightMcp
             ? [...allowedPermissionRules, PLAYWRIGHT_MCP_ALLOWED_TOOL_RULE]
             : allowedPermissionRules;
+    // --safe-mode disables ALL MCP servers, including ones passed explicitly via --mcp-config
+    // (verified empirically on claude 2.1.207). Playwright judge sessions therefore drop it:
+    // their CLAUDE_CONFIG_DIR is an isolated per-task directory with no user customizations,
+    // so the surface --safe-mode guards is already empty, while --strict-mcp-config and the
+    // permission rules below stay in force.
     const args = [
         '-p',
-        '--safe-mode',
+        ...(input.playwrightMcp ? [] : ['--safe-mode']),
         '--strict-mcp-config',
         '--model',
         model,

package/dist/commands/worker.d.ts

@@ -137,6 +137,8 @@ export declare function resolveWorkerClaimTaskAssignment(claim: WorkerClaimAgent
 export declare function nextClaimBackoffMs(currentMs: number): number;
 export declare function claimBackoffDelayMs(currentMs: number, random?: () => number): number;
 export declare function resolveWorkerHomeDir(): string;
+export declare const WORKER_TASK_WORKSPACE_RETENTION = 20;
+export declare function pruneWorkerTaskWorkspaces(incomingWorkspaceDir: string, retention?: number): Promise<string[]>;
 export declare function hasAgentTaskUploadedArtifact(task: Pick<AgentTaskResponse, 'appId' | 'appVersionId'>): boolean;
 export declare function appendWorkerTranscriptChunks(input: {
     client: Pick<ApiClient, 'workerAppendAgentTaskTranscriptChunks'>;

package/dist/commands/worker.js

@@ -3,7 +3,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.WORKER_CONTEXT_COMMAND_NOT_ALLOWED_MESSAGE = exports.WORKER_SESSION_EXPIRED_MESSAGE = exports.runLoggedProcess = exports.readEnvBoolean = exports.readCodexSandboxMode = exports.readCodexRolloutTokenUsageForThread = exports.extractCodexTokensUsed = exports.extractCodexThreadId = exports.extractCodexRolloutTokenUsage = exports.extractAgentTokenUsage = exports.DEFAULT_WORKER_TOKEN_CAP = exports.DEFAULT_CODEX_TIMEOUT_MS = exports.buildWorkerChildEnv = exports.buildCodexExecArgs = exports.buildClaudePermissionSettings = exports.buildClaudeExecArgs = exports.assertWorkerTokenUsageWithinCap = void 0;
+exports.WORKER_TASK_WORKSPACE_RETENTION = exports.WORKER_CONTEXT_COMMAND_NOT_ALLOWED_MESSAGE = exports.WORKER_SESSION_EXPIRED_MESSAGE = exports.runLoggedProcess = exports.readEnvBoolean = exports.readCodexSandboxMode = exports.readCodexRolloutTokenUsageForThread = exports.extractCodexTokensUsed = exports.extractCodexThreadId = exports.extractCodexRolloutTokenUsage = exports.extractAgentTokenUsage = exports.DEFAULT_WORKER_TOKEN_CAP = exports.DEFAULT_CODEX_TIMEOUT_MS = exports.buildWorkerChildEnv = exports.buildCodexExecArgs = exports.buildClaudePermissionSettings = exports.buildClaudeExecArgs = exports.assertWorkerTokenUsageWithinCap = void 0;
 exports.allocateWorkerDevPort = allocateWorkerDevPort;
 exports.isWorkerContextCommandAllowed = isWorkerContextCommandAllowed;
 exports.resolveWorkerExecutionTargetFromRole = resolveWorkerExecutionTargetFromRole;
@@ -11,6 +11,7 @@ exports.resolveWorkerClaimTaskAssignment = resolveWorkerClaimTaskAssignment;
 exports.nextClaimBackoffMs = nextClaimBackoffMs;
 exports.claimBackoffDelayMs = claimBackoffDelayMs;
 exports.resolveWorkerHomeDir = resolveWorkerHomeDir;
+exports.pruneWorkerTaskWorkspaces = pruneWorkerTaskWorkspaces;
 exports.hasAgentTaskUploadedArtifact = hasAgentTaskUploadedArtifact;
 exports.appendWorkerTranscriptChunks = appendWorkerTranscriptChunks;
 exports.readTaskNextStepsFile = readTaskNextStepsFile;
@@ -412,6 +413,33 @@ function workerStateFilePath() {
 function workerTaskWorkspaceDirFromAssignment(assignment) {
     return node_path_1.default.join(resolveWorkerHomeDir(), 'tasks', assignment.workspace.folderName);
 }
+exports.WORKER_TASK_WORKSPACE_RETENTION = 20;
+// Task workspaces (including retained failed ones) otherwise grow without bound on a long-lived
+// worker host. Keep only the most recently touched ones; the incoming task's own directory is
+// never a prune candidate.
+async function pruneWorkerTaskWorkspaces(incomingWorkspaceDir, retention = exports.WORKER_TASK_WORKSPACE_RETENTION) {
+    const tasksRoot = node_path_1.default.dirname(node_path_1.default.resolve(incomingWorkspaceDir));
+    if (!(0, node_fs_1.existsSync)(tasksRoot)) {
+        return [];
+    }
+    const incoming = node_path_1.default.resolve(incomingWorkspaceDir);
+    const entries = (0, node_fs_1.readdirSync)(tasksRoot, { withFileTypes: true })
+        .filter((entry) => entry.isDirectory())
+        .map((entry) => node_path_1.default.join(tasksRoot, entry.name))
+        .filter((dir) => dir !== incoming)
+        .map((dir) => ({ dir, mtimeMs: (0, node_fs_1.statSync)(dir).mtimeMs }))
+        .sort((a, b) => b.mtimeMs - a.mtimeMs);
+    const stale = entries.slice(Math.max(0, retention - 1));
+    const removed = [];
+    for (const entry of stale) {
+        await (0, promises_1.rm)(entry.dir, { recursive: true, force: true });
+        removed.push(entry.dir);
+    }
+    if (removed.length > 0) {
+        console.error(`pruned ${removed.length} stale task workspace(s), keeping the ${retention} most recent`);
+    }
+    return removed;
+}
 function buildWorkerTaskContext(claim, assignment) {
     const task = claim.task;
     if (!task) {
@@ -3870,6 +3898,7 @@ async function startWorker(options = {}) {
                     reasoningEffort = codexModel?.reasoningEffort ?? claudeModel?.effort ?? null;
                     const taskContext = buildWorkerTaskContext(claim, assignment);
                     workspaceDir = workerTaskWorkspaceDirFromAssignment(assignment);
+                    await pruneWorkerTaskWorkspaces(workspaceDir);
                     await (0, promises_1.rm)(workspaceDir, { recursive: true, force: true });
                     await (0, promises_1.mkdir)(workspaceDir, { recursive: true });
                     const eventDir = node_path_1.default.join(workspaceDir, '.playdrop-task-events');

package/node_modules/@playdrop/config/client-meta.json

@@ -1,6 +1,6 @@
 {
   "version": "0.12.8",
-  "build": 9,
+  "build": 11,
   "runtimeSdkVersion": "0.12.5",
   "runtimeSdkBuild": 2,
   "clients": {

package/node_modules/@playdrop/types/dist/instrument-evidence.d.ts

@@ -28,7 +28,9 @@ export type InstrumentEvidenceManifest = {
 };
 export declare const PLAYWRIGHT_MCP_SCREENSHOT_TOOL = "mcp__playwright__browser_take_screenshot";
 export declare const PLAYWRIGHT_MCP_CONTROLLED_TAB_ID = 1;
+export declare const PLAYWRIGHT_MCP_SCREENSHOT_BUDGET = 100;
 export declare function isInstrumentScreenshotId(value: string): boolean;
+export declare function normalizePlaywrightScreenshotId(value: unknown): string | null;
 export declare class ClaudeChromeScreenshotStreamParser {
     private readonly onScreenshot;
     private buffer;
@@ -44,9 +46,9 @@ export declare class ClaudeChromeScreenshotStreamParser {
 export declare class PlaywrightMcpScreenshotStreamParser {
     private readonly onScreenshot;
     private buffer;
-    private readonly pendingScreenshotToolUseIds;
+    private readonly pendingScreenshotFilenames;
     private readonly handledToolUseIds;
-    private sequence;
+    private screenshotCallCount;
     constructor(onScreenshot: (screenshot: ClaudeChromeScreenshot) => void);
     push(chunk: string): number;
     finish(): number;

package/node_modules/@playdrop/types/dist/instrument-evidence.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"instrument-evidence.d.ts","sourceRoot":"","sources":["../src/instrument-evidence.ts"],"names":[],"mappings":"AAAA,eAAO,MAAM,yBAAyB,iDAK5B,CAAC;AAEX,MAAM,MAAM,sBAAsB,GAAG,CAAC,OAAO,yBAAyB,CAAC,CAAC,MAAM,CAAC,CAAC;AAEhF,MAAM,MAAM,sBAAsB,GAAG;IACnC,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,YAAY,GAAG,WAAW,CAAC;IACtC,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;CACvB,CAAC;AAEF,MAAM,MAAM,yBAAyB,GAAG;IACtC,IAAI,EAAE,sBAAsB,CAAC;IAC7B,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,YAAY,GAAG,WAAW,CAAC;IACtC,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;CACpB,CAAC;AAEF,MAAM,MAAM,0BAA0B,GAAG;IACvC,OAAO,EAAE,CAAC,CAAC;IACX,OAAO,EAAE,MAAM,CAAC;IAChB,eAAe,EAAE,MAAM,CAAC;IACxB,QAAQ,EAAE,OAAO,CAAC,MAAM,CAAC,sBAAsB,EAAE,yBAAyB,CAAC,CAAC,CAAC;CAC9E,CAAC;AAMF,eAAO,MAAM,8BAA8B,6CAA6C,CAAC;AACzF,eAAO,MAAM,gCAAgC,IAAI,CAAC;AAGlD,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAE/D;AA+CD,qBAAa,kCAAkC;IAKjC,OAAO,CAAC,QAAQ,CAAC,YAAY;IAJzC,OAAO,CAAC,MAAM,CAAM;IACpB,OAAO,CAAC,QAAQ,CAAC,yBAAyB,CAA+B;IACzE,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAAqB;gBAE7B,YAAY,EAAE,CAAC,UAAU,EAAE,sBAAsB,KAAK,IAAI;IAEvF,IAAI,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM;IAqB3B,MAAM,IAAI,MAAM;IAMhB,OAAO,CAAC,WAAW;IA0BnB,OAAO,CAAC,aAAa;IAkBrB,OAAO,CAAC,iBAAiB;CAiE1B;AAMD,qBAAa,mCAAmC;IAMlC,OAAO,CAAC,QAAQ,CAAC,YAAY;IALzC,OAAO,CAAC,MAAM,CAAM;IACpB,OAAO,CAAC,QAAQ,CAAC,2BAA2B,CAAqB;IACjE,OAAO,CAAC,QAAQ,CAAC,iBAAiB,CAAqB;IACvD,OAAO,CAAC,QAAQ,CAAK;gBAEQ,YAAY,EAAE,CAAC,UAAU,EAAE,sBAAsB,KAAK,IAAI;IAEvF,IAAI,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM;IAqB3B,MAAM,IAAI,MAAM;IAMhB,OAAO,CAAC,WAAW;IA0BnB,OAAO,CAAC,aAAa;IAcrB,OAAO,CAAC,iBAAiB;CA8C1B"}
+{"version":3,"file":"instrument-evidence.d.ts","sourceRoot":"","sources":["../src/instrument-evidence.ts"],"names":[],"mappings":"AAAA,eAAO,MAAM,yBAAyB,iDAK5B,CAAC;AAEX,MAAM,MAAM,sBAAsB,GAAG,CAAC,OAAO,yBAAyB,CAAC,CAAC,MAAM,CAAC,CAAC;AAEhF,MAAM,MAAM,sBAAsB,GAAG;IACnC,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,YAAY,GAAG,WAAW,CAAC;IACtC,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;CACvB,CAAC;AAEF,MAAM,MAAM,yBAAyB,GAAG;IACtC,IAAI,EAAE,sBAAsB,CAAC;IAC7B,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,YAAY,GAAG,WAAW,CAAC;IACtC,QAAQ,EAAE,MAAM,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;CACpB,CAAC;AAEF,MAAM,MAAM,0BAA0B,GAAG;IACvC,OAAO,EAAE,CAAC,CAAC;IACX,OAAO,EAAE,MAAM,CAAC;IAChB,eAAe,EAAE,MAAM,CAAC;IACxB,QAAQ,EAAE,OAAO,CAAC,MAAM,CAAC,sBAAsB,EAAE,yBAAyB,CAAC,CAAC,CAAC;CAC9E,CAAC;AAMF,eAAO,MAAM,8BAA8B,6CAA6C,CAAC;AACzF,eAAO,MAAM,gCAAgC,IAAI,CAAC;AAGlD,eAAO,MAAM,gCAAgC,MAAM,CAAC;AAKpD,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAE/D;AAED,wBAAgB,+BAA+B,CAAC,KAAK,EAAE,OAAO,GAAG,MAAM,GAAG,IAAI,CAG7E;AA+CD,qBAAa,kCAAkC;IAKjC,OAAO,CAAC,QAAQ,CAAC,YAAY;IAJzC,OAAO,CAAC,MAAM,CAAM;IACpB,OAAO,CAAC,QAAQ,CAAC,yBAAyB,CAA+B;IACzE,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAAqB;gBAE7B,YAAY,EAAE,CAAC,UAAU,EAAE,sBAAsB,KAAK,IAAI;IAEvF,IAAI,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM;IAqB3B,MAAM,IAAI,MAAM;IAMhB,OAAO,CAAC,WAAW;IA0BnB,OAAO,CAAC,aAAa;IAkBrB,OAAO,CAAC,iBAAiB;CAiE1B;AAMD,qBAAa,mCAAmC;IAMlC,OAAO,CAAC,QAAQ,CAAC,YAAY;IALzC,OAAO,CAAC,MAAM,CAAM;IACpB,OAAO,CAAC,QAAQ,CAAC,0BAA0B,CAA6B;IACxE,OAAO,CAAC,QAAQ,CAAC,iBAAiB,CAAqB;IACvD,OAAO,CAAC,mBAAmB,CAAK;gBAEH,YAAY,EAAE,CAAC,UAAU,EAAE,sBAAsB,KAAK,IAAI;IAEvF,IAAI,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM;IAqB3B,MAAM,IAAI,MAAM;IAMhB,OAAO,CAAC,WAAW;IA0BnB,OAAO,CAAC,aAAa;IAwBrB,OAAO,CAAC,iBAAiB;CA+C1B"}

package/node_modules/@playdrop/types/dist/instrument-evidence.js

@@ -20,9 +20,11 @@ __export(instrument_evidence_exports, {
   ClaudeChromeScreenshotStreamParser: () => ClaudeChromeScreenshotStreamParser,
   INSTRUMENT_EVIDENCE_NAMES: () => INSTRUMENT_EVIDENCE_NAMES,
   PLAYWRIGHT_MCP_CONTROLLED_TAB_ID: () => PLAYWRIGHT_MCP_CONTROLLED_TAB_ID,
+  PLAYWRIGHT_MCP_SCREENSHOT_BUDGET: () => PLAYWRIGHT_MCP_SCREENSHOT_BUDGET,
   PLAYWRIGHT_MCP_SCREENSHOT_TOOL: () => PLAYWRIGHT_MCP_SCREENSHOT_TOOL,
   PlaywrightMcpScreenshotStreamParser: () => PlaywrightMcpScreenshotStreamParser,
-  isInstrumentScreenshotId: () => isInstrumentScreenshotId
+  isInstrumentScreenshotId: () => isInstrumentScreenshotId,
+  normalizePlaywrightScreenshotId: () => normalizePlaywrightScreenshotId
 });
 module.exports = __toCommonJS(instrument_evidence_exports);
 const INSTRUMENT_EVIDENCE_NAMES = [
@@ -36,10 +38,15 @@ const SCREENSHOT_ID_IN_TEXT_PATTERN = /\bID:\s*(ss_[A-Za-z0-9_-]+)\b/g;
 const MAX_STREAM_LINE_LENGTH = 20 * 1024 * 1024;
 const PLAYWRIGHT_MCP_SCREENSHOT_TOOL = "mcp__playwright__browser_take_screenshot";
 const PLAYWRIGHT_MCP_CONTROLLED_TAB_ID = 1;
-const PLAYWRIGHT_SCREENSHOT_ID_PATTERN = /^pw-[1-9][0-9]*$/;
+const PLAYWRIGHT_MCP_SCREENSHOT_BUDGET = 100;
+const PLAYWRIGHT_SCREENSHOT_ID_PATTERN = /^[a-z0-9][a-z0-9._-]{0,63}\.(png|jpe?g)$/;
 function isInstrumentScreenshotId(value) {
   return SCREENSHOT_ID_PATTERN.test(value) || PLAYWRIGHT_SCREENSHOT_ID_PATTERN.test(value);
 }
+function normalizePlaywrightScreenshotId(value) {
+  const id = typeof value === "string" ? value.trim().toLowerCase() : "";
+  return PLAYWRIGHT_SCREENSHOT_ID_PATTERN.test(id) ? id : null;
+}
 function asRecord(value) {
   return value && typeof value === "object" && !Array.isArray(value) ? value : null;
 }
@@ -221,9 +228,9 @@ class ClaudeChromeScreenshotStreamParser {
 class PlaywrightMcpScreenshotStreamParser {
   onScreenshot;
   buffer = "";
-  pendingScreenshotToolUseIds = /* @__PURE__ */ new Set();
+  pendingScreenshotFilenames = /* @__PURE__ */ new Map();
   handledToolUseIds = /* @__PURE__ */ new Set();
-  sequence = 0;
+  screenshotCallCount = 0;
   constructor(onScreenshot) {
     this.onScreenshot = onScreenshot;
   }
@@ -287,7 +294,16 @@ class PlaywrightMcpScreenshotStreamParser {
       if (item?.["type"] !== "tool_use" || !id || name !== PLAYWRIGHT_MCP_SCREENSHOT_TOOL) {
         continue;
       }
-      this.pendingScreenshotToolUseIds.add(id);
+      this.screenshotCallCount += 1;
+      if (this.screenshotCallCount > PLAYWRIGHT_MCP_SCREENSHOT_BUDGET) {
+        throw new Error(`playwright_screenshot_budget_exceeded:${PLAYWRIGHT_MCP_SCREENSHOT_BUDGET}`);
+      }
+      const input = asRecord(item?.["input"]);
+      const filename = normalizePlaywrightScreenshotId(input?.["filename"]);
+      if (!filename) {
+        continue;
+      }
+      this.pendingScreenshotFilenames.set(id, filename);
     }
   }
   recordToolResults(payload) {
@@ -300,11 +316,12 @@ class PlaywrightMcpScreenshotStreamParser {
         continue;
       }
       const toolUseId = typeof toolResult["tool_use_id"] === "string" ? toolResult["tool_use_id"].trim() : "";
-      if (!this.pendingScreenshotToolUseIds.has(toolUseId) || this.handledToolUseIds.has(toolUseId)) {
+      const screenshotId = this.pendingScreenshotFilenames.get(toolUseId);
+      if (!screenshotId || this.handledToolUseIds.has(toolUseId)) {
         continue;
       }
       this.handledToolUseIds.add(toolUseId);
-      this.pendingScreenshotToolUseIds.delete(toolUseId);
+      this.pendingScreenshotFilenames.delete(toolUseId);
       const resultContent = Array.isArray(toolResult["content"]) ? toolResult["content"] : [];
       const images = resultContent.map(asRecord).filter((item) => item?.["type"] === "image");
       if (images.length === 0) {
@@ -319,9 +336,8 @@ class PlaywrightMcpScreenshotStreamParser {
       if (mediaType !== "image/jpeg" && mediaType !== "image/png" || !dataBase64) {
         throw new Error(`invalid_playwright_screenshot_payload:${toolUseId}`);
       }
-      this.sequence += 1;
       this.onScreenshot({
-        screenshotId: `pw-${this.sequence}`,
+        screenshotId,
         tabId: PLAYWRIGHT_MCP_CONTROLLED_TAB_ID,
         toolUseId,
         mediaType,
@@ -339,7 +355,9 @@ class PlaywrightMcpScreenshotStreamParser {
   ClaudeChromeScreenshotStreamParser,
   INSTRUMENT_EVIDENCE_NAMES,
   PLAYWRIGHT_MCP_CONTROLLED_TAB_ID,
+  PLAYWRIGHT_MCP_SCREENSHOT_BUDGET,
   PLAYWRIGHT_MCP_SCREENSHOT_TOOL,
   PlaywrightMcpScreenshotStreamParser,
-  isInstrumentScreenshotId
+  isInstrumentScreenshotId,
+  normalizePlaywrightScreenshotId
 });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@playdrop/playdrop-cli",
-  "version": "0.12.16",
+  "version": "0.12.18",
   "description": "Official Playdrop CLI for publishing browser games, creator apps, and AI-generated game assets on playdrop.ai",
   "homepage": "https://www.playdrop.ai",
   "repository": {