@blamejs/exceptd-skills 0.13.100 → 0.13.102

package/CHANGELOG.md

@@ -1,5 +1,13 @@
 # Changelog
 
+## 0.13.102 — 2026-05-25
+
+CVE catalog — prompt injection to code execution in natural-language data-analysis agents. Adds two flaws in agents whose purpose is to turn a natural-language question into code that the framework then runs, so prompt injection is the exploit primitive. **CVE-2024-5565** (Vanna.AI, CWE-94 / CWE-77, JFrog CNA CVSS 8.1 HIGH; GitHub advisory 9.2; NVD unscored) — the text-to-SQL `ask` method runs LLM-generated Python to build a Plotly visualization (default-on), so an injected question executes arbitrary Python on the host. **CVE-2024-12366** (PandasAI, CWE-94, CISA-ADP CVSS 9.8 CRITICAL; NVD unscored) — the `chat` interface runs LLM-generated Python against DataFrames without separating analytical input from injected instructions, giving unauthenticated RCE / sandbox escape. Neither has a fixed release, so both are scored without patch credit and remediation is sandboxing the code-execution path; both map MITRE ATLAS AML.T0051 (LLM Prompt Injection) and ATT&CK T1059.006, and introduce a control (NEW-CTRL-102) requiring NL-to-code/SQL agents to treat the question and analyzed data as untrusted and never run model-generated code with host privileges. CVE count 381 → 383.
+
+## 0.13.101 — 2026-05-25
+
+CVE catalog — vector-database RCE and backup path traversal. Adds two more flaws in the RAG persistence layer. **CVE-2026-45829** (ChromaDB "ChromaToast", CWE-94, CNA CVSS v4.0 10.0 CRITICAL; NVD unscored) — ChromaDB's Python FastAPI server processes a caller-supplied embedding-function config (a model repo with `trust_remote_code=true`) on the collections endpoint *before* authenticating, giving unauthenticated remote code execution; no fixed Python release is published, so mitigation is network isolation, the Rust `chroma run` / official Docker deployment, and disabling remote model loading. **CVE-2025-67818** (Weaviate, CWE-22, NIST CVSS 7.2) — backup restore does not constrain entry paths, so a write-capable attacker uses absolute / `../` paths (ZipSlip) to create or overwrite arbitrary host files; fixed in 1.33.4. Both map MITRE ATLAS AML.T0049 and ATT&CK T1190; ChromaDB reuses the vector-DB authentication control (NEW-CTRL-101) shared with Milvus, and Weaviate reuses the path-traversal control (NEW-CTRL-094) shared with the Ollama / AnythingLLM entries. The unpatched pre-auth RCE scores well above the patched path-traversal flaw under RWEP. CVE count 379 → 381.
+
 ## 0.13.100 — 2026-05-25
 
 CVE catalog — PyTorch torch.load RCE despite weights_only=True. Adds **CVE-2025-32434** (CWE-502, NIST CVSS 9.8 CRITICAL): PyTorch's `torch.load` executes attacker code from a crafted checkpoint even when called with `weights_only=True` — the setting the ecosystem recommended as the safe way to load untrusted models — so pipelines that followed that guidance on ≤ 2.5.1 remain vulnerable; fixed in 2.6.0. Maps MITRE ATLAS AML.T0010 / AML.T0011 / AML.T0011.000 and ATT&CK T1204 / T1059 / T1195.002, and reuses the untrusted-model-artifact control (NEW-CTRL-091) shared with the Keras, Hugging Face Transformers, and NeMo entries — a model checkpoint is executable code regardless of "safe" load flags. CVE count 378 → 379.

package/data/_indexes/_meta.json

@@ -1,21 +1,21 @@
 {
   "schema_version": "1.1.0",
-  "generated_at": "2026-05-26T05:42:13.652Z",
+  "generated_at": "2026-05-26T06:46:06.333Z",
   "generator": "scripts/build-indexes.js",
   "source_count": 54,
   "source_hashes": {
-    "manifest.json": "a124f4c5277f994a7b103df4168494c8966a7bcb6ec4b6b0b80923cad2473e95",
-    "data/atlas-ttps.json": "0c3b467b6f2f522506340ecd9f72192940475dbf2a8fcb59db967b5d9cf7dadd",
-    "data/attack-techniques.json": "643203bd90b130cbb4a6ec06197532057a5951bb38058df12d7ebdcd6f66a1c7",
-    "data/cve-catalog.json": "71080698bd88b7f5a9e6a5f8e544a2501945f2dd5b5a8db3796d02802c060da5",
-    "data/cwe-catalog.json": "8a66b854f7e80c5b11dd53e80d7acc79671346145fc4f9bfc9f01842005185fc",
+    "manifest.json": "94703e7d93b8acb54cc44ea919df56dc5d2dd47d8741a775cd2879f8485decc1",
+    "data/atlas-ttps.json": "95682d60e815e4cce2ba2e973b7f47b4f1d11d0bbdcb1ef720f2ca5947dc5ba8",
+    "data/attack-techniques.json": "452a40aa2876ccc28791f3b643e018c38c72167a91e2a4123ecda2d7b6a4f6bb",
+    "data/cve-catalog.json": "16f203c85ed6a003b9bac9543b041d67d4bfcc7ea19c1239651101b08411c1a8",
+    "data/cwe-catalog.json": "97033c785065cf314a64f0c614f0e86dbfd53be466c5580a53d36bc6a099c3e2",
     "data/d3fend-catalog.json": "9a54bccb9f24f84b32024216cc3f53819a053721ac8ab43c326859e68fc0ffaf",
     "data/dlp-controls.json": "d2406c482dddd30e49203879999dc4b3a7fd4d0494d6a61d86b91ee76415df19",
     "data/exploit-availability.json": "ec2656f0d9a893610e27b43eb6035fe9b18e057c9f6dfaac7e7d4959bbcbb795",
-    "data/framework-control-gaps.json": "8e906ae7533754388aff17a6d84bec20413c39d764540622acf36570e9f235c4",
+    "data/framework-control-gaps.json": "841949de4006de3b72c0a1d56850c10c124293197a8f9c9ce4cbdb61545df9e3",
     "data/global-frameworks.json": "9ba563a85f7f8d6c3c957de64945e20925a89d0ed6ea6fc561cf093811acf558",
     "data/rfc-references.json": "66ef2e1f444a2cf0c2700a754f0a66030bb8a91d9e68394b9537ea1fe8b904fe",
-    "data/zeroday-lessons.json": "293979e529dc05909e330705ec5086abf36cc541fc153329d7b24d7c6bf5ead6",
+    "data/zeroday-lessons.json": "fe65f9fafddb07614eb5c520b899a1b44d620b4be982c088362a1d4f45ce6588",
     "skills/kernel-lpe-triage/skill.md": "08b3e9815ba481c57c80f5fc0ccbf5bb7cbb41f570c235ba6ff9596b8c07354d",
     "skills/ai-attack-surface/skill.md": "c4c1eb22a38ca7a959b5725222bab8fbd4f4044a548a93f3e288e6f698334b72",
     "skills/mcp-agent-trust/skill.md": "89ac89084391d2341b6513fefb1be2d36b93de1c130f057696219c1c59440f13",
@@ -72,7 +72,7 @@
       "dlp_refs": 0
     },
     "trigger_table_entries": 538,
-    "chains_cve_entries": 368,
+    "chains_cve_entries": 372,
     "chains_cwe_entries": 171,
     "jurisdictions_indexed": 29,
     "handoff_dag_nodes": 42,

package/data/_indexes/activity-feed.json

@@ -149,7 +149,7 @@
       "artifact": "data/cve-catalog.json",
       "path": "data/cve-catalog.json",
       "schema_version": "1.0.0",
-      "entry_count": 379
+      "entry_count": 383
     },
     {
       "date": "2026-05-18",
@@ -165,7 +165,7 @@
       "artifact": "data/zeroday-lessons.json",
       "path": "data/zeroday-lessons.json",
       "schema_version": "1.1.0",
-      "entry_count": 374
+      "entry_count": 378
     },
     {
       "date": "2026-05-17",

package/data/_indexes/catalog-summaries.json

@@ -62,7 +62,7 @@
         "rebuild_after_days": 365,
         "note": "Per-entry last_verified governs decay. Skills depending on this catalog must check entry freshness before high-stakes use."
       },
-      "entry_count": 379,
+      "entry_count": 383,
       "sample_keys": [
         "CVE-2025-53773",
         "CVE-2026-30615",
@@ -238,7 +238,7 @@
         "rebuild_after_days": 365,
         "note": "Per-entry last_verified governs decay. Skills depending on this catalog must check entry freshness before high-stakes use."
       },
-      "entry_count": 374,
+      "entry_count": 378,
       "sample_keys": [
         "CVE-2026-31431",
         "CVE-2025-53773",