npm - @runtypelabs/sdk - Versions diffs - 1.7.3 → 1.8.0 - Mend

@runtypelabs/sdk 1.7.3 → 1.8.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (7) hide show

package/dist/index.cjs CHANGED Viewed

@@ -327,6 +327,7 @@ __export(index_exports, {
   defaultWorkflow: () => defaultWorkflow,
   deployWorkflow: () => deployWorkflow,
   evaluateGeneratedRuntimeToolProposal: () => evaluateGeneratedRuntimeToolProposal,
+  gameWorkflow: () => gameWorkflow,
   getDefaultPlanPath: () => getDefaultPlanPath,
   getLikelySupportingCandidatePaths: () => getLikelySupportingCandidatePaths,
   isDiscoveryToolName: () => isDiscoveryToolName,
@@ -2909,8 +2910,8 @@ var scaffoldPhase = {
     ].join("\n");
   },
   interceptToolCall(toolName, _args, _ctx) {
-    const blockedTools = ["write_file", "read_file", "search_repo", "glob_files", "tree_directory", "list_directory", "restore_file_checkpoint"];
-    if (blockedTools.includes(toolName)) {
+    const blockedTools2 = ["write_file", "read_file", "search_repo", "glob_files", "tree_directory", "list_directory", "restore_file_checkpoint"];
+    if (blockedTools2.includes(toolName)) {
       return [
         `Blocked: ${toolName} is not available in deploy mode.`,
         "Use deploy_sandbox to deploy your code to a live sandbox instead of writing files to the local repo."
@@ -2949,6 +2950,7 @@ var deployPhase = {
       "",
       "deploy_sandbox takes:",
       "  - code: The full source code (written to main.ts/main.js/main.py)",
+      '  - files: Additional files (path \u2192 content), e.g. { "public/index.html": "<html>..." }',
       '  - packageJson: Dependencies as a JSON object, e.g. { "dependencies": { "express": "^4.18.2" } }',
       '  - language: "typescript" (default), "javascript", or "python"',
       "  - port: The port your server listens on (default: 3000)",
@@ -2962,7 +2964,8 @@ var deployPhase = {
       "  3. Include ALL dependencies in packageJson \u2014 nothing is pre-installed except Node.js built-ins.",
       "  4. If the deploy fails, read the error output, fix the code, and call deploy_sandbox again.",
       "  5. The sandbox is persistent \u2014 subsequent calls reuse the same sandbox.",
-      "  6. When the deploy succeeds and the preview URL is live, tell the user and end with TASK_COMPLETE."
+      "  6. When the deploy succeeds and the preview URL is live, tell the user and end with TASK_COMPLETE.",
+      "  7. For apps with HTML frontends, use the `files` parameter to write HTML/CSS/JS to separate files and serve them with express.static, rather than embedding HTML in template literals."
     ].join("\n");
   },
   buildToolGuidance(_state) {
@@ -2977,8 +2980,8 @@ var deployPhase = {
     return false;
   },
   interceptToolCall(toolName, _args, _ctx) {
-    const blockedTools = ["write_file", "read_file", "search_repo", "glob_files", "tree_directory", "list_directory", "restore_file_checkpoint"];
-    if (blockedTools.includes(toolName)) {
+    const blockedTools2 = ["write_file", "read_file", "search_repo", "glob_files", "tree_directory", "list_directory", "restore_file_checkpoint"];
+    if (blockedTools2.includes(toolName)) {
       return [
         `Blocked: ${toolName} is not available in deploy mode.`,
         "Use deploy_sandbox to deploy your code to a live sandbox instead of writing files to the local repo."
@@ -2986,8 +2989,13 @@ var deployPhase = {
     }
     return void 0;
   },
-  canAcceptCompletion(_state, trace) {
-    return trace.entries.some((entry) => entry.startsWith("deploy_sandbox"));
+  canAcceptCompletion(state, trace) {
+    if (trace.entries.some((entry) => entry.startsWith("deploy_sandbox"))) {
+      return true;
+    }
+    return state.sessions.some(
+      (s) => s.actionKeys?.some((key) => key.startsWith("deploy_sandbox"))
+    );
   },
   buildRecoveryMessage(state) {
     const recent = state.sessions.slice(-2);
@@ -3058,6 +3066,204 @@ var deployWorkflow = {
   }
 };
+// src/workflows/game-workflow.ts
+var blockedTools = ["write_file", "read_file", "search_repo", "glob_files", "tree_directory", "list_directory", "restore_file_checkpoint"];
+function blockLocalTools(toolName) {
+  if (blockedTools.includes(toolName)) {
+    return [
+      `Blocked: ${toolName} is not available in game deploy mode.`,
+      "Use deploy_sandbox to deploy your game to a live sandbox instead of writing files to the local repo."
+    ].join(" ");
+  }
+  return void 0;
+}
+var designPhase = {
+  name: "design",
+  description: "Understand game requirements and plan the approach",
+  buildInstructions(_state) {
+    return [
+      "--- Workflow Phase: Design ---",
+      "This is a game development task. You will build a game and deploy it to a live sandbox \u2014 you are NOT editing files in the local repository.",
+      "Quickly determine:",
+      "  1. What type of game (3D, 2D, platformer, racing, puzzle, etc.)",
+      "  2. Theme and visual style",
+      "  3. Core game mechanics (controls, scoring, win/lose conditions)",
+      "  4. Tech stack (Three.js, Phaser, vanilla Canvas/WebGL, etc.)",
+      "  5. Any specific dependencies needed",
+      "",
+      "Do NOT inspect the local repo or search for files to edit.",
+      "Do NOT write a plan file.",
+      "Once you understand the requirements, proceed to build and deploy."
+    ].join("\n");
+  },
+  buildToolGuidance(_state) {
+    return [
+      "This is a sandbox deploy task \u2014 do NOT use write_file, read_file, search_repo, glob_files, tree_directory, or list_directory.",
+      "Your primary tool is deploy_sandbox. Use it to deploy code to a persistent sandbox and get a live preview URL."
+    ];
+  },
+  isComplete(ctx) {
+    return ctx.state.sessions.length >= 1 || ctx.trace.entries.some((entry) => entry.startsWith("deploy_sandbox"));
+  },
+  buildTransitionSummary(_state, _nextPhaseName) {
+    return [
+      "Automatic phase transition: design \u2192 build.",
+      "Requirements understood. Build the game and deploy it using deploy_sandbox with the files parameter."
+    ].join("\n");
+  },
+  interceptToolCall(toolName, _args, _ctx) {
+    return blockLocalTools(toolName);
+  },
+  shouldForceEndTurn(snapshot, _ctx) {
+    if (snapshot.pauseCount >= 12) {
+      return "design phase is looping without progressing \u2014 end the turn so the system can advance to build";
+    }
+    return void 0;
+  }
+};
+var buildPhase = {
+  name: "build",
+  description: "Build the game and deploy to sandbox with live preview",
+  buildInstructions(_state) {
+    return [
+      "--- Workflow Phase: Build ---",
+      "Build the game and deploy it using the deploy_sandbox tool.",
+      "",
+      "CRITICAL: Multi-file deployment pattern",
+      "  You MUST use the `files` parameter for the game HTML/JS/CSS.",
+      "  NEVER embed HTML inside a JavaScript template literal in Express `res.send()` \u2014 this WILL break",
+      "  because game code uses backticks (template literals) which cause nested backtick corruption.",
+      "",
+      "  Correct pattern:",
+      "    - code: A minimal Express static file server (~5 lines):",
+      "      ```",
+      '      const express = require("express");',
+      "      const app = express();",
+      '      app.use(express.static("public"));',
+      '      app.listen(3000, () => console.log("Server running on port 3000"));',
+      "      ```",
+      "    - files: Your game assets as separate files:",
+      "      {",
+      '        "public/index.html": "<!DOCTYPE html>...",',
+      '        "public/game.js": "// game logic...",',
+      '        "public/style.css": "body { margin: 0; }"',
+      "      }",
+      "",
+      "  WRONG pattern (DO NOT DO THIS):",
+      '    - code: `app.get("/", (req, res) => res.send(\\`<html>...game code with backticks...\\`))`',
+      "    - This breaks because game code uses template literals inside the res.send template literal.",
+      "",
+      "deploy_sandbox takes:",
+      "  - code: The server source code (written to main.ts/main.js/main.py)",
+      '  - files: Additional files (path \u2192 content), e.g. { "public/index.html": "<html>..." }',
+      '  - packageJson: Dependencies as a JSON object, e.g. { "dependencies": { "express": "^4.18.2", "three": "^0.170.0" } }',
+      '  - language: "javascript" (recommended for games), "typescript", or "python"',
+      "  - port: The port your server listens on (default: 3000)",
+      "  - startCommand: Custom start command (auto-detected by default)",
+      "",
+      "Guidelines:",
+      '  1. Use `language: "javascript"` \u2014 simpler for static file servers.',
+      "  2. The `code` param should be a minimal Express static server. ALL game code goes in `files`.",
+      "  3. Include ALL dependencies in packageJson (express, three, phaser, etc.).",
+      "  4. If the deploy fails, read the error output, fix the code, and call deploy_sandbox again.",
+      "  5. The sandbox is persistent \u2014 subsequent calls reuse the same sandbox.",
+      "  6. Load game libraries from CDN in your HTML (e.g., Three.js from unpkg/cdnjs) OR include them in packageJson.",
+      "  7. Make the game fullscreen by default (width: 100vw, height: 100vh, no margin/padding on body)."
+    ].join("\n");
+  },
+  buildToolGuidance(_state) {
+    return [
+      "Your primary tool is deploy_sandbox. Call it with code (Express static server), files (game HTML/JS/CSS), and packageJson.",
+      "ALWAYS use the `files` parameter for game HTML, JavaScript, and CSS \u2014 NEVER embed them in template literals.",
+      "Do NOT use write_file, read_file, search_repo, glob_files, tree_directory, or list_directory.",
+      "You may use run_sandbox_code for quick one-off tests if needed."
+    ];
+  },
+  isComplete() {
+    return false;
+  },
+  interceptToolCall(toolName, _args, _ctx) {
+    return blockLocalTools(toolName);
+  },
+  canAcceptCompletion(_state, trace) {
+    return trace.entries.some((entry) => entry.startsWith("deploy_sandbox"));
+  },
+  buildRecoveryMessage(state) {
+    const recent = state.sessions.slice(-2);
+    if (recent.length < 2) return void 0;
+    const noProgress = recent.every(
+      (s) => s.hadTextOutput === false && s.wroteFiles === false
+    );
+    if (!noProgress) return void 0;
+    return [
+      "Recovery instruction:",
+      "You should be deploying the game using deploy_sandbox with the `files` parameter.",
+      "Use `code` for a minimal Express static server and `files` for game HTML/JS/CSS.",
+      "NEVER embed HTML in a JavaScript template literal \u2014 use separate files."
+    ].join("\n");
+  },
+  shouldForceEndTurn(snapshot, _ctx) {
+    if (snapshot.consecutiveDiscoveryPauseCount >= 8) {
+      return "build phase is looping on discovery tools instead of calling deploy_sandbox";
+    }
+    if (snapshot.actionKeyCount >= 4) {
+      return `the same action repeated ${snapshot.actionKeyCount} times \u2014 try a different approach`;
+    }
+    return void 0;
+  }
+};
+var verifyPhase = {
+  name: "verify",
+  description: "Verify the game is running and playable",
+  buildInstructions(_state) {
+    return [
+      "--- Workflow Phase: Verify ---",
+      "The game has been deployed. Verify it is working:",
+      "  1. Check the server log for errors. If there are errors, fix and redeploy using deploy_sandbox.",
+      "  2. Tell the user the preview URL so they can play the game.",
+      "  3. End with TASK_COMPLETE.",
+      "",
+      "If the user reports issues, fix the code and redeploy."
+    ].join("\n");
+  },
+  buildToolGuidance(_state) {
+    return [
+      "Use deploy_sandbox to redeploy if fixes are needed.",
+      "Do NOT use local file system tools.",
+      "When the game is working, tell the user the preview URL and end with TASK_COMPLETE."
+    ];
+  },
+  isComplete() {
+    return false;
+  },
+  interceptToolCall(toolName, _args, _ctx) {
+    return blockLocalTools(toolName);
+  },
+  canAcceptCompletion(_state, trace) {
+    return trace.entries.some((entry) => entry.startsWith("deploy_sandbox"));
+  },
+  shouldForceEndTurn(snapshot, _ctx) {
+    if (snapshot.actionKeyCount >= 4) {
+      return `the same action repeated ${snapshot.actionKeyCount} times \u2014 try a different approach`;
+    }
+    return void 0;
+  }
+};
+function classifyVariant3() {
+  return "game";
+}
+var gameWorkflow = {
+  name: "game",
+  phases: [designPhase, buildPhase, verifyPhase],
+  classifyVariant: classifyVariant3,
+  async generateBootstrapContext() {
+    return void 0;
+  },
+  buildCandidateBlock() {
+    return "";
+  }
+};
 // src/endpoints.ts
 var FlowsEndpoint = class {
   constructor(client) {
@@ -5313,13 +5519,15 @@ var _AgentsEndpoint = class _AgentsEndpoint {
         maxSessions,
         localToolNames,
         continuationContext,
-        workflow
+        workflow,
+        options.toolIds
       );
       let sessionResult;
       const sessionData = {
         messages,
         debugMode: options.debugMode,
-        model: options.model
+        model: options.model,
+        ...options.toolIds?.length ? { tools: { toolIds: options.toolIds } } : {}
       };
       let sessionToolMessages = [];
       if (useStream && options.localTools) {
@@ -5343,6 +5551,7 @@ var _AgentsEndpoint = class _AgentsEndpoint {
           result: completeEvent.finalOutput || "",
           iterations: completeEvent.iterations,
           totalCost: completeEvent.totalCost || 0,
+          totalTokens: completeEvent.totalTokens,
           stopReason: completeEvent.stopReason,
           error: completeEvent.error
         };
@@ -5360,6 +5569,7 @@ var _AgentsEndpoint = class _AgentsEndpoint {
           result: completeEvent.finalOutput || "",
           iterations: completeEvent.iterations,
           totalCost: completeEvent.totalCost || 0,
+          totalTokens: completeEvent.totalTokens,
           stopReason: completeEvent.stopReason,
           error: completeEvent.error
         };
@@ -5372,8 +5582,15 @@ var _AgentsEndpoint = class _AgentsEndpoint {
         toolTraceSummary
       );
       const sessionCost = sessionResult.totalCost;
+      const sessionTokens = sessionResult.totalTokens;
       state.sessionCount = session + 1;
       state.totalCost += sessionCost;
+      if (sessionTokens) {
+        state.totalTokens = {
+          input: (state.totalTokens?.input || 0) + sessionTokens.input,
+          output: (state.totalTokens?.output || 0) + sessionTokens.output
+        };
+      }
       state.lastOutput = effectiveSessionOutput;
       state.lastError = sessionResult.stopReason === "error" ? sessionResult.error || "Agent session ended with an error." : void 0;
       state.lastStopReason = sessionResult.stopReason;
@@ -5381,6 +5598,7 @@ var _AgentsEndpoint = class _AgentsEndpoint {
       state.sessions.push({
         index: session + 1,
         cost: sessionCost,
+        totalTokens: sessionTokens,
         iterations: sessionResult.iterations,
         stopReason: sessionResult.stopReason,
         outputPreview: effectiveSessionOutput.slice(0, 300),
@@ -5494,6 +5712,7 @@ var _AgentsEndpoint = class _AgentsEndpoint {
       status: state.status,
       sessionCount: state.sessionCount,
       totalCost: state.totalCost,
+      totalTokens: state.totalTokens,
       lastOutput: state.lastOutput,
       sessions: state.sessions,
       recordId
@@ -5540,12 +5759,12 @@ var _AgentsEndpoint = class _AgentsEndpoint {
    * Build messages for a session, injecting progress context for continuation sessions.
    * Optionally accepts continuation context for marathon resume scenarios.
    */
-  buildSessionMessages(originalMessage, state, sessionIndex, maxSessions, localToolNames, continuationContext, workflow) {
+  buildSessionMessages(originalMessage, state, sessionIndex, maxSessions, localToolNames, continuationContext, workflow, builtinToolIds) {
     const wf = workflow ?? defaultWorkflow;
     const currentPhase = wf.phases.find((p) => p.name === state.workflowPhase);
     const toolGuidanceLines = currentPhase?.buildToolGuidance(state) ?? [];
     const isDeployWorkflow = wf.name === "deploy";
-    const toolsBlock = localToolNames?.length ? [
+    const localToolsBlock = localToolNames?.length ? [
       "",
       "--- Local Tools ---",
       `You have access to tools (${localToolNames.join(", ")}) that execute directly on the user's machine.`,
@@ -5555,6 +5774,14 @@ var _AgentsEndpoint = class _AgentsEndpoint {
       ...toolGuidanceLines,
       ...isDeployWorkflow ? [] : ["Always use write_file to save your output so the user can run it immediately."]
     ].join("\n") : "";
+    const builtinToolNames = builtinToolIds?.map((id) => id.replace(/^builtin:/, ""));
+    const builtinToolsBlock = builtinToolNames?.length ? [
+      "",
+      "--- Built-in Tools ---",
+      `You have access to built-in tools (${builtinToolNames.join(", ")}) for web search, web scraping, image generation, and other capabilities.`,
+      "Use these tools when the task requires information from the web, generating images, or other capabilities beyond local file operations."
+    ].join("\n") : "";
+    const toolsBlock = localToolsBlock + builtinToolsBlock;
     const bootstrapBlock = state.bootstrapContext ? ["", "--- Initial Repository Discovery ---", state.bootstrapContext].join("\n") : "";
     const phaseBlock = ["", this.buildPhaseInstructions(state, wf)].join("\n");
     const candidateBlock = wf.buildCandidateBlock?.(state) ?? "";
@@ -7011,6 +7238,7 @@ var ClientEvalBuilder = class extends EvalBuilder {
   defaultWorkflow,
   deployWorkflow,
   evaluateGeneratedRuntimeToolProposal,
+  gameWorkflow,
   getDefaultPlanPath,
   getLikelySupportingCandidatePaths,
   isDiscoveryToolName,