update theagent to work with docker

frontend/src/utils/status.ts

@@ -37,10 +37,8 @@ export const INITIAL_AGENT_STATES: AgentState[] = [
     color: "#3fb950",
     status: "pending",
     steps: [
-      { id: "test.oracle", label: "Preparando scaffold", status: "pending" },
-      { id: "test.gen", label: "Gerando PoC", status: "pending" },
-      { id: "test.run", label: "Executando Foundry", status: "pending" },
-      { id: "test.reflect", label: "Analisando falha", status: "pending" },
+      { id: "test.gen", label: "Codificando PoC (LLM)", status: "pending" },
+      { id: "test.run", label: "Executando Sandbox (Ferramentas)", status: "pending" },
     ],
   },
 ];

src/agents/tester/graph.ts

@@ -27,8 +27,12 @@ function routeAfterAgent(state: PoCState): "pocoToolsNode" | typeof END {
   return END;
 }
 
+import { emitStep } from "../../logger.js";
+
 // A simple node to update the toolCallCount after tools run
 function trackToolCallsNode(state: PoCState): Partial<PoCState> {
+  emitStep({ agent: "tester", step: "run", status: "running" });
+
   const messages = state.messages;
   const lastMessage = messages[messages.length - 1];
   
@@ -39,6 +43,11 @@ function trackToolCallsNode(state: PoCState): Partial<PoCState> {
     }
   }
 
+  if (newStatus === "success") {
+    emitStep({ agent: "tester", step: "run", status: "done" });
+    emitStep({ agent: "tester", step: "gen", status: "done" });
+  }
+
   return {
     toolCallCount: 1, // reducer is additive (+1)
     status: newStatus,

src/agents/tester/nodes/context.ts

@@ -1,6 +1,7 @@
 import fs from "fs/promises";
 import path from "path";
 import { PoCState } from "../state.js";
+import { logger } from "../../../logger.js";
 
 export async function contextNode(state: PoCState): Promise<Partial<PoCState>> {
   console.log("[contextNode] Preparando ambiente de testes para:", state.report.title);

src/agents/tester/nodes/pocoAgent.ts

@@ -2,6 +2,7 @@ import { HumanMessage, SystemMessage, AIMessage, trimMessages } from "@langchain
 import { PoCState } from "../state.js";
 import { pocoTools } from "../tools.js";
 import { createLLM } from "../../../config/llm.js";
+import { emitStep } from "../../../logger.js";
 
 const MAX_STEPS = 30; // Max tool calls threshold
 const MAX_COST_USD = 3.0; // Max cost threshold
@@ -43,6 +44,7 @@ function calculateCost(inputTokens: number, outputTokens: number): number {
 }
 
 export async function pocoAgentNode(state: PoCState): Promise<Partial<PoCState>> {
+  emitStep({ agent: "tester", step: "gen", status: "running" });
   let messages = state.messages || [];
 
   // Check limits
@@ -126,8 +128,9 @@ export async function pocoAgentNode(state: PoCState): Promise<Partial<PoCState>>
   }
 
   return {
-    messages: [...initialMessages, response],
+    messages: [response],
     totalCost: runCost,
-    iterations: 1, // Add 1 to total iterations tracking
+    toolCallCount: 1, // Reducer is additive
+    iterations: 1, // Reducer is additive
   };
 }

src/agents/tester/tools.ts

@@ -186,7 +186,7 @@ export const smartContractCompileTool = tool(
         { 
           cwd: sandboxDir, 
           timeout: 30000,
-          env: { ...process.env, PATH: `${process.env.HOME}/.foundry/bin:${process.env.PATH}` } 
+          env: { ...process.env } 
         }
       );
       
@@ -220,7 +220,7 @@ export const smartContractTestTool = tool(
         { 
           cwd: sandboxDir, 
           timeout: 60000,
-          env: { ...process.env, PATH: `${process.env.HOME}/.foundry/bin:${process.env.PATH}` } 
+          env: { ...process.env } 
         }
       );
       

src/benchmark/runSyntheticEvaluation.ts

@@ -38,9 +38,9 @@ function appendCsvRow(row: string[]) {
 async function runEvaluation() {
   const cases = await parseJSONL(JSONL_FILE);
   
-  // Only evaluate 10 hard cases to match the amount of easy/intermediate cases
-  const targetCases = cases.filter(c => c.complexity === "hard").slice(0, 10);
-  console.log(`Iniciando avaliação para ${targetCases.length} projetos sintéticos difíceis...`);
+  // Roda a avaliação para os datasets easy e intermediate a pedido do usuário
+  const targetCases = cases.filter(c => c.complexity === "easy" || c.complexity === "intermediate");
+  console.log(`Iniciando avaliação para ${targetCases.length} projetos sintéticos (easy/intermediate)...`);
 
   const csvHeaders = [
     "Task_ID",

src/server.ts

@@ -1,6 +1,6 @@
 import "dotenv/config";
 
-import { mkdirSync, writeFileSync } from "node:fs";
+import { mkdirSync, writeFileSync, readFileSync } from "node:fs";
 import { resolve } from "node:path";
 import { tmpdir } from "node:os";
 import { serve } from "@hono/node-server";
@@ -13,6 +13,7 @@ import { coderAgent } from "./agents/coder/agent.ts";
 import { auditorAgent } from "./agents/auditor/agent.ts";
 import { testerAgent } from "./agents/tester/agent.ts";
 import { mapFindingToReport } from "./utils/mapFinding.js";
+import { createEmptyFoundryProject } from "./utils/forgeSandbox.js";
 import { logger, setLogSink, clearLogSink, setStepSink, clearStepSink } from "./logger.ts";
 
 const app = new Hono();
@@ -87,18 +88,41 @@ app.post("/api/run", (c) => {
           coderResult.contract,
           auditorResult.repoContext   // ← now forwarded to tester
         );
-        report.customSandboxDir = outputDir;  // ← tester runs in real project sandbox
-        const testerResult = await testerAgent.invoke({ report });
+        const sandboxDir = resolve(tmpdir(), `talp1-tester-${Date.now()}`);
+        await createEmptyFoundryProject(sandboxDir, coderResult.contract, "Contract");
+        report.customSandboxDir = sandboxDir;  // ← tester runs in real project sandbox
+        report.affectedContract.sourceFilePath = "src/Contract.sol"; // Fix bug with relative path
+
+        const testerResult = await testerAgent.invoke(
+          { report },
+          { recursionLimit: 100, configurable: { sandboxDir } }
+        );
 
         logger.info(`[Tester] Execução concluída com status: ${testerResult.status}`);
 
+        let finalPocCode = testerResult.pocCode || "";
+        try {
+          finalPocCode = readFileSync(resolve(sandboxDir, "test/Exploit.t.sol"), "utf-8");
+        } catch (e) {
+          // Ignorar se não criou
+        }
+
+        let finalExecutionLogs: string[] = testerResult.executionLogs || [];
+        if (finalExecutionLogs.length === 0 && testerResult.messages) {
+          const testMsgs = testerResult.messages.filter((m: any) => m._getType() === "tool" && m.name === "smart_contract_test");
+          if (testMsgs.length > 0) {
+            const content = testMsgs[testMsgs.length - 1].content;
+            finalExecutionLogs = [typeof content === "string" ? content : JSON.stringify(content)];
+          }
+        }
+
         // Garante que o objeto enviado tem exatamente o que o front espera
         await send(
           "tester",
           JSON.stringify({
             status: testerResult.status,
-            pocCode: testerResult.pocCode,
-            executionLogs: testerResult.executionLogs,
+            pocCode: finalPocCode,
+            executionLogs: finalExecutionLogs,
             iterations: testerResult.iterations,
           }),
         );

src/utils/forgeSandbox.ts

@@ -6,7 +6,7 @@ export async function createEmptyFoundryProject(targetDir: string, sourceCode: s
   await fs.mkdir(targetDir, { recursive: true });
   execSync("forge init --no-git --force", { 
     cwd: targetDir,
-    env: { ...process.env, PATH: `${process.env.PATH}:/home/tales/.foundry/bin` }
+    env: { ...process.env } // Deixa o PATH nativo agir (configurado pelo bash ou Dockerfile)
   });
   
   // Clean up default files