pre-release 0.1.4

.github/workflows/docker-build.yml

@@ -1,15 +1,18 @@
+# Pushes to Docker Hub only on: push to main/release_* (not on pull_request), or workflow_dispatch.
+# PRs build the image for CI but intentionally do not push (no registry credentials / avoid polluting tags).
+
 name: Build and Push Deadend Docker Image
 
 on:
   push:
     branches: [ main, release_* ]
     paths:
-      - 'setup/images/kalilinux.Dockerfile'
+      - 'environments/images/kalilinux.Dockerfile'
       - '.github/workflows/docker-build.yml'
   pull_request:
     branches: [ main ]
     paths:
-      - 'setup/images/kalilinux.Dockerfile'
+      - 'environments/images/kalilinux.Dockerfile'
       - '.github/workflows/docker-build.yml'
   workflow_dispatch:
     inputs:
@@ -20,7 +23,7 @@ on:
 
 env:
   REGISTRY: docker.io
-  IMAGE_NAME: bargacy/deadend-pentest
+  IMAGE_NAME: xoxruns/sandboxed_kali
 
 jobs:
   build-and-push:
@@ -36,6 +39,22 @@ jobs:
     - name: Set up Docker Buildx
       uses: docker/setup-buildx-action@v3
 
+    - name: Explain no push on pull requests
+      if: github.event_name == 'pull_request'
+      run: |
+        echo "PR workflow: image is built only (push=false). After merge to main or release_*, a push event will build and push to Docker Hub."
+
+    - name: Require Docker Hub secrets when pushing
+      if: github.event_name != 'pull_request'
+      env:
+        DOCKER_USER: ${{ secrets.DOCKER_USERNAME }}
+        DOCKER_PASS: ${{ secrets.DOCKER_TOKEN }}
+      run: |
+        if [ -z "$DOCKER_USER" ] || [ -z "$DOCKER_PASS" ]; then
+          echo "::error::Add repository Actions secrets DOCKER_USERNAME and DOCKER_TOKEN (Docker Hub access token). Without them, the job cannot log in and nothing is pushed to Docker Hub."
+          exit 1
+        fi
+
     - name: Log in to Docker Hub
       if: github.event_name != 'pull_request'
       uses: docker/login-action@v3
@@ -58,10 +77,11 @@ jobs:
           type=raw,value=${{ github.event.inputs.tag }},enable=${{ github.event.inputs.tag != '' }}
 
     - name: Build and push Docker image
+      id: build
       uses: docker/build-push-action@v5
       with:
         context: .
-        file: ./setup/images/kalilinux.Dockerfile
+        file: ./environments/images/kalilinux.Dockerfile
         push: ${{ github.event_name != 'pull_request' }}
         tags: ${{ steps.meta.outputs.tags }}
         labels: ${{ steps.meta.outputs.labels }}

.gitmodules

@@ -4,3 +4,6 @@
 [submodule "benchmarks/xbow/validation-benchmarks-xbow"]
 	path = benchmarks/xbow/validation-benchmarks-xbow
 	url = https://github.com/xoxruns/validation-benchmarks-xbow.git
+[submodule "deadend_cli/simple-python-interpreter-sandbox"]
+	path = deadend_cli/simple-python-interpreter-sandbox
+	url = https://github.com/xoxruns/simple-python-interpreter-sandbox.git

benchmarks/run_xbow_benchmark.sh

@@ -248,7 +248,7 @@ echo "[+] Launching eval agent with uv run"
 echo "[+] Logging uv run output to: $LOG_FILE"
 (
     cd "$REPO_ROOT/deadend_cli/src/deadend_cli" && \
-    uv run main.py eval-agent --eval-metadata-file "$META_FILE" --llm-providers azure_ai
+    uv run main.py eval-agent --eval-metadata-file "$META_FILE" --provider azure_ai --model-name Kimi-K2.5
 ) 2>&1 | tee "$LOG_FILE"
 
 echo "[+] Stopping benchmark services with make stop"

cli/deadend_cli/components/ComponentHealth.tsx

@@ -22,7 +22,7 @@ export interface ComponentHealthProps {
 // Short display names for components
 const COMPONENT_NAMES: Record<string, string> = {
   docker: "Docker",
-  pgvector: "pgvector",
+  rag: "RAG",
   config: "Config",
   python_sandbox: "Python",
   shell_sandbox: "Shell",
@@ -31,8 +31,8 @@ const COMPONENT_NAMES: Record<string, string> = {
 // Order to display components
 const COMPONENT_ORDER = [
   "docker",
-  "pgvector",
   "config",
+  "rag",
   "python_sandbox",
   "shell_sandbox",
 ];

cli/deadend_cli/hooks/useComponentHealth.ts

@@ -4,7 +4,7 @@ import type { DeadEndRpcClient } from "../runtime/deadend-rpc-client.ts";
 
 export type ComponentName =
   | "docker"
-  | "pgvector"
+  | "rag"
   | "config"
   | "python_sandbox"
   | "shell_sandbox";
@@ -104,8 +104,8 @@ export function useComponentHealth(
           case "docker":
             result = await rpcClient.initDocker();
             break;
-          case "pgvector":
-            result = await rpcClient.initPgvector();
+          case "rag":
+            result = await rpcClient.initRag();
             break;
           case "config":
             result = await rpcClient.initConfig();
@@ -154,8 +154,8 @@ export function useComponentHealth(
 
     const components: ComponentName[] = [
       "docker",
-      "pgvector",
       "config",
+      "rag",
       "python_sandbox",
       "shell_sandbox",
     ];

cli/deadend_cli/main.tsx

@@ -109,7 +109,7 @@ function App({ cliArgs }: AppProps) {
 
         // Initialize all components at once using init_all
         setInitStatus("Initializing all components...");
-        // We can wait longer here for pgvector and the sandbox
+        // Allow time for sandboxes and Playwright during init_all
         const initResult = await client.initAll(300000);
 
         // Store component results for display
@@ -125,7 +125,7 @@ function App({ cliArgs }: AppProps) {
         }
 
         // Check for critical failures (all components required for task execution)
-        const criticalComponents = ["docker", "config", "model_registry", "pgvector", "shell_sandbox"];
+        const criticalComponents = ["docker", "config", "model_registry", "rag", "shell_sandbox"];
         const criticalFailures = initResult.failed_components.filter(
           (c) => criticalComponents.includes(c)
         );

cli/deadend_cli/runtime/deadend-rpc-client.ts

@@ -24,7 +24,7 @@
  * │  ┌─────────────────────────────────────────────────────────────────────────┐│
  * │  │                        DeadEndRpcClient                                 ││
  * │  │  - runTask() / runTaskWithCallbacks()                                   ││
- * │  │  - healthAll() / initDocker() / initPgvector() / ...                    ││
+ * │  │  - healthAll() / initDocker() / initRag() / ...                         ││
  * │  │  - subscribeEvents() / interrupt() / approve()                          ││
  * │  └────────────────────────────────────────────────────────────────────────┘│
  * │                                │                                            │
@@ -67,7 +67,7 @@
  *
  * // Initialize components
  * await client.initDocker();
- * await client.initPgvector();
+ * await client.initRag();
  * await client.initShellSandbox();
  *
  * // Run security testing task
@@ -301,8 +301,8 @@ export interface DeadEndRpcClientOptions extends StdioRpcClientOptions {
  *
  * Before running tasks, components must be initialized in order:
  * 1. `initDocker()` - Docker daemon connection (required)
- * 2. `initPgvector()` - Vector database for RAG (optional)
- * 3. `initConfig()` - Load LLM API keys and settings
+ * 2. `initConfig()` - Load LLM API keys and settings
+ * 3. `initRag()` - SQLite-backed RAG session manager (after config, for storage paths)
  * 4. `initShellSandbox()` - Prepare Kali container for shell commands
  * 5. `initPythonSandbox()` - Start Python interpreter sandbox
  * 6. `initPlaywright()` - Browser automation (optional)
@@ -443,7 +443,7 @@ export class DeadEndRpcClient {
    *
    * Returns a comprehensive health report including:
    * - Docker daemon connectivity
-   * - pgvector database status
+   * - RAG (SQLite) session manager status
    * - Python sandbox process status
    * - Shell sandbox readiness
    * - Playwright browser status
@@ -466,12 +466,12 @@ export class DeadEndRpcClient {
   }
 
   /**
-   * Checks pgvector database health.
+   * Checks RAG (SQLite) session manager health.
    *
-   * @returns Promise resolving to HealthResult for pgvector
+   * @returns Promise resolving to HealthResult for RAG
    */
-  async healthPgvector(): Promise<HealthResult> {
-    const result = await this.client.call("health_pgvector");
+  async healthRag(): Promise<HealthResult> {
+    const result = await this.client.call("health_rag");
     return result as HealthResult;
   }
 
@@ -523,17 +523,14 @@ export class DeadEndRpcClient {
   }
 
   /**
-   * Initializes the pgvector database container.
+   * Initializes the SQLite-backed RAG session manager.
    *
-   * Starts the pgvector container if not running and verifies
-   * database connectivity. Used for RAG (retrieval-augmented generation).
-   *
-   * Requires: initDocker() must be called first
+   * Prefer calling after `initConfig()` so storage paths from config apply.
    *
    * @returns Promise resolving to InitResult with success status
    */
-  async initPgvector(): Promise<InitResult> {
-    const result = await this.client.call("init_pgvector");
+  async initRag(): Promise<InitResult> {
+    const result = await this.client.call("init_rag");
     return result as InitResult;
   }
 
@@ -611,9 +608,9 @@ export class DeadEndRpcClient {
    * proper dependency order and provides a comprehensive result.
    *
    * Initialization order:
-   * 1. Docker (required by pgvector and shell_sandbox)
-   * 2. Config (required by model_registry)
-   * 3. pgvector (requires Docker)
+   * 1. Docker (required by shell_sandbox)
+   * 2. Config (required by model_registry and RAG paths)
+   * 3. RAG session manager (SQLite, no Docker)
    * 4. Model Registry (requires Config)
    * 5. Python sandbox (standalone)
    * 6. Shell sandbox (requires Docker)
@@ -1093,8 +1090,7 @@ export class DeadEndRpcClient {
    * - Playwright browser
    * - Python sandbox process
    * - Shell sandbox containers
-   * - pgvector database (optional)
-   * - RAG connector
+   * - RAG session manager (SQLite)
    *
    * @returns Promise resolving to shutdown status for each component
    */

cli/deadend_cli/types/rpc.ts

@@ -501,12 +501,10 @@ export interface AgentErrorData {
  *
  * @property thought - The full reasoning text
  * @property summary - Condensed version of the thought
- * @property relevance - How relevant this thought is to the task (0.0-1.0)
  */
 export interface AgentThoughtData {
   thought: string;
   summary?: string;
-  relevance: number;
 }
 
 /**
@@ -647,7 +645,7 @@ export type ComponentStatus =
  *
  * Returned by health_* RPC methods to report component status.
  *
- * @property component - Name of the component (docker, pgvector, etc.)
+ * @property component - Name of the component (docker, rag, config, etc.)
  * @property healthy - Whether the component is functioning correctly
  * @property status - Current lifecycle state of the component
  * @property message - Human-readable status message