chore(patterns): add SUP-029 Strapi npm malware + PSV-009 Langflow Agentic RCE [2026.04.06.1]

PATTERN_UPDATES.md

@@ -1,3 +1,14 @@
+## 2026-04-06
+rulepack: 2026.04.06.1
+Two new detection rules, IOC enrichment, and vuln DB updates.
+- Added `SUP-029` (critical): **Malicious Strapi npm packages (Redis RCE / Credential Harvesting)** — 36 malicious npm packages disguised as Strapi CMS plugins were discovered in April 2026. These packages use postinstall scripts to exploit Redis and PostgreSQL, deploy reverse shells, harvest credentials, and drop persistent implants.
+- Added `PSV-009` (critical): **Langflow Agentic Assistant RCE Vulnerability (CVE-2026-33873)** — A remote code execution flaw in Langflow Agentic Assistant allows attackers to execute LLM-generated Python code server-side, potentially leading to system compromise.
+- IOC update: added `prod-strapi` to domain IOC DB.
+- Vuln DB update: added `langflow` python version 1.0.0 as CVE-2026-33873 (critical, fixed 1.0.1).
+Sources:
+- The Hacker News (Strapi): https://thehackernews.com/2026/04/36-malicious-npm-packages-exploited.html
+- GBHackers (Strapi): https://gbhackers.com/36-malicious-strapi-npm/
+- SentinelOne (Langflow): https://www.sentinelone.com/vulnerability-database/cve-2026-33873/
 ## 2026-04-05
 rulepack: 2026.04.05.1
 

examples/showcase/147_sup029_strapi_npm_malware/SKILL.md

@@ -0,0 +1,12 @@
+---
+name: strapi-plugin-nordica-tools
+description: A malicious Strapi plugin that deploys Redis RCE and harvests credentials.
+---
+
+# Strapi Plugin Nordica Tools
+
+This plugin provides advanced tools for Strapi CMS.
+
+```bash
+npm install strapi-plugin-nordica-tools
+```

examples/showcase/148_psv009_langflow_agentic_rce/SKILL.md

@@ -0,0 +1,14 @@
+---
+name: langflow-agentic-rce-test
+description: A test skill for Langflow Agentic Assistant RCE (CVE-2026-33873).
+---
+
+# Langflow Agentic RCE Test
+
+This skill tests the Langflow Agentic Assistant RCE vulnerability.
+
+```python
+# CVE-2026-33873
+import langflow
+# agentic rce
+```

examples/showcase/INDEX.md

@@ -168,3 +168,5 @@ skillscan scan examples/showcase/29_claude_sed_path_bypass --fail-on never
 146. `146_sup027_telnyx_version`: expects `SUP-027`
 147. `100_nomshub_cursor_tunnel_sandbox_escape`: NomShub cursor-tunnel sandbox escape and persistence via shell builtins (`MAL-061`)
 148. `101_unc1069_npm_maintainer_social_engineering`: UNC1069 social engineering lure domain teams.onlivemeet.com targeting npm maintainers (`SUP-028`)
+149. `147_sup029_strapi_npm_malware`: Malicious Strapi npm packages (Redis RCE / Credential Harvesting) (`SUP-029`)
+150. `148_psv009_langflow_agentic_rce`: Langflow Agentic Assistant RCE Vulnerability (CVE-2026-33873) (`PSV-009`)

scripts/adversarial_eval.py

@@ -31,10 +31,9 @@
 import random
 import re
 import sys
-from datetime import datetime, timezone
+from collections.abc import Callable
+from datetime import UTC, datetime
 from pathlib import Path
-from typing import Callable
-
 
 # ---------------------------------------------------------------------------
 # Perturbation techniques
@@ -231,17 +230,16 @@ def load_model(model_path: str, tokenizer_dir: str) -> tuple[object, object]:
         import onnxruntime as ort
         from transformers import AutoTokenizer
     except ImportError as exc:
-        sys.exit(
-            f"Missing dependency: {exc}\n"
-            "Install with: pip install skillscan-security[ml-onnx]"
-        )
+        sys.exit(f"Missing dependency: {exc}\nInstall with: pip install skillscan-security[ml-onnx]")
 
     session = ort.InferenceSession(model_path, providers=["CPUExecutionProvider"])
     tokenizer = AutoTokenizer.from_pretrained(tokenizer_dir)
     return session, tokenizer
 
 
-def predict_injection_probability(session: object, tokenizer: object, text: str, max_length: int = 384) -> float:
+def predict_injection_probability(
+    session: object, tokenizer: object, text: str, max_length: int = 384
+) -> float:
     """Return injection probability (0–1) for the given text."""
     import numpy as np
 
@@ -374,7 +372,10 @@ def main(argv: list[str] | None = None) -> int:
         for name in args.techniques.split(","):
             name = name.strip().lower()
             if name not in _ALL_TECHNIQUES:
-                print(f"ERROR: unknown technique '{name}'. Choose from: {', '.join(_ALL_TECHNIQUES)}", file=sys.stderr)
+                print(
+                    f"ERROR: unknown technique '{name}'. Choose from: {', '.join(_ALL_TECHNIQUES)}",
+                    file=sys.stderr,
+                )
                 return 1
             selected[name] = _ALL_TECHNIQUES[name]
 
@@ -407,7 +408,7 @@ def main(argv: list[str] | None = None) -> int:
         "n_injection_samples": len(samples),
         "techniques": results,
         "overall_vulnerable": any_vulnerable,
-        "generated_at": datetime.now(timezone.utc).isoformat(),
+        "generated_at": datetime.now(UTC).isoformat(),
         "threshold_asr": 0.10,
     }
 

scripts/append_rules.py

@@ -0,0 +1,117 @@
+import sys
+from datetime import datetime
+
+
+def main():
+    file_path = "src/skillscan/data/rules/default.yaml"
+    with open(file_path) as f:
+        lines = f.readlines()
+
+    # Find the index of capability_patterns:
+    insert_idx = -1
+    for i, line in enumerate(lines):
+        if line.startswith("capability_patterns:"):
+            insert_idx = i
+            break
+
+    if insert_idx == -1:
+        print("Could not find capability_patterns:")
+        sys.exit(1)
+
+    # Update version
+    for i, line in enumerate(lines):
+        if line.startswith("version:"):
+            # e.g. version: "2026.04.05.1" -> "2026.04.06.1"
+            today = datetime.now().strftime("%Y.%m.%d")
+            if today in line:
+                # increment the last digit
+                parts = line.strip().split(".")
+                last = int(parts[-1].replace('"', ""))
+                lines[i] = f'version: "{today}.{last + 1}"\n'
+            else:
+                lines[i] = f'version: "{today}.1"\n'
+            break
+
+    new_rules = """  - id: SUP-029
+    category: supply_chain
+    severity: critical
+    confidence: 0.95
+    title: Malicious Strapi npm packages (Redis RCE / Credential Harvesting)
+    pattern: (?i)(?:strapi-plugin-(?:cron|config|server|database|core|hooks|monitor|events|logger|health|sync|seed|locale|form|notify|api|sitemap-gen|nordica-tools|nordica-sync|nordica-cms|nordica-api|nordica-recon|nordica-stage|nordica-vhost|nordica-deep|nordica-lite|nordica|finseven|hextest|cms-tools|content-sync|debug-tools|health-check|guardarian-ext|advanced-uuid|blurhash))
+    mitigation: '36 malicious npm packages disguised as Strapi CMS plugins were discovered in April 2026. These packages use postinstall scripts to exploit Redis and PostgreSQL, deploy reverse shells, harvest credentials, and drop persistent implants. Remove the package immediately and rotate all credentials, especially Redis, PostgreSQL, and cryptocurrency wallet keys.'
+    metadata:
+      version: 1.0.0
+      status: active
+      author: skillscan
+      created: '2026-04-06'
+      updated: '2026-04-06'
+      tags:
+        - supply_chain
+        - npm
+        - strapi
+        - redis
+        - rce
+        - credential-harvesting
+      last_modified: '2026-04-06'
+      techniques:
+        - id: SUP-029
+          name: Supply Chain Compromise
+      applies_to:
+        contexts:
+          - source
+          - workflow
+      lifecycle:
+        introduced: '2026-04-06'
+        last_modified: '2026-04-06'
+      quality:
+        precision_estimate: 0.95
+        benchmark_set: core-static-v1
+      references:
+        - https://thehackernews.com/2026/04/36-malicious-npm-packages-exploited.html
+        - https://gbhackers.com/36-malicious-strapi-npm/
+  - id: PSV-009
+    category: vulnerability
+    severity: critical
+    confidence: 0.95
+    title: Langflow Agentic Assistant RCE Vulnerability (CVE-2026-33873)
+    pattern: (?i)(?:CVE-2026-33873|langflow[^\n]{0,200}agentic[^\n]{0,200}rce)
+    mitigation: 'A remote code execution flaw in Langflow Agentic Assistant (CVE-2026-33873) allows attackers to execute LLM-generated Python code server-side, potentially leading to system compromise. Update Langflow to a patched version.'
+    metadata:
+      version: 1.0.0
+      status: active
+      author: skillscan
+      created: '2026-04-06'
+      updated: '2026-04-06'
+      tags:
+        - vulnerability
+        - langflow
+        - rce
+        - cve-2026-33873
+      last_modified: '2026-04-06'
+      techniques:
+        - id: PSV-009
+          name: Vulnerability Exploitation
+      applies_to:
+        contexts:
+          - source
+          - workflow
+      lifecycle:
+        introduced: '2026-04-06'
+        last_modified: '2026-04-06'
+      quality:
+        precision_estimate: 0.95
+        benchmark_set: core-static-v1
+      references:
+        - https://www.sentinelone.com/vulnerability-database/cve-2026-33873/
+"""
+
+    lines.insert(insert_idx, new_rules)
+
+    with open(file_path, "w") as f:
+        f.writelines(lines)
+
+    print("Rules appended successfully.")
+
+
+if __name__ == "__main__":
+    main()

scripts/generate_examples_table.py

@@ -15,7 +15,7 @@
 for yf in sorted(rules_dir.glob("*.yaml")):
     with open(yf) as f:
         data = yaml.safe_load(f)
-    for r in data.get("rules", []):
+    for r in data.get("static_rules", data.get("rules", [])):
         r["_source"] = yf.name
         all_rules.append(r)
     for r in data.get("chain_rules", []):
@@ -38,6 +38,7 @@
     "obfuscation": "Obfuscation & Evasion",
     "defense_evasion": "Defense Evasion",
     "uncategorized": "Other",
+    "vulnerability": "Vulnerabilities",
 }
 
 lines = [

scripts/verify_onnx_export.py

@@ -87,8 +87,7 @@ def _run_pt_forward(model: torch.nn.Module, enc: dict) -> np.ndarray:
         # ── Step 3: PyTorch forward passes on 5 test strings ─────────────────
         print("[3/8] Running PyTorch forward passes on 5 test strings…")
         encodings = [
-            tokenizer(text, return_tensors="pt", truncation=True, max_length=512)
-            for text in TEST_STRINGS
+            tokenizer(text, return_tensors="pt", truncation=True, max_length=512) for text in TEST_STRINGS
         ]
         pt_outputs = [_run_pt_forward(model, enc) for enc in encodings]
         print(f"      output shape per sample: {pt_outputs[0].shape}")
@@ -183,10 +182,7 @@ def _run_pt_forward(model: torch.nn.Module, enc: dict) -> np.ndarray:
         latencies_ms.sort()
         report["median_inference_ms"] = round(latencies_ms[49], 3)
         report["p95_inference_ms"] = round(latencies_ms[94], 3)
-        print(
-            f"      Median: {report['median_inference_ms']} ms  "
-            f"P95: {report['p95_inference_ms']} ms"
-        )
+        print(f"      Median: {report['median_inference_ms']} ms  P95: {report['p95_inference_ms']} ms")
 
     except Exception as exc:
         print(f"\nERROR: {exc}", file=sys.stderr)