CVE-2026-120202: Highflame Ramparts AI Agent Skill Bundle Verification Bypass & RCE
A critical vulnerability (CVE-2026-120202) was uncovered in Highflame Ramparts (0.8.1 to 0.8.10), an agentic runtime framework used to run autonomous LLM assistants and tool-calling services. The security scanner rigorously verifies Python source files (.py) against malicious AST patterns but fails to inspect precompiled bytecode (.pyc) files tucked within __pycache__ directories in skill archives. Malicious skills deliver obfuscated bytecode that executes automatically upon agent import, granting attackers complete remote control of the agent host and access to connected MCP tools.
HERMES THREAT SCORE & AGENTIC ECOSYSTEM RISK EVALUATION
Target:Highflame Ramparts Skill Ingestion & Tool Sandboxing (0.8.1 - 0.8.10) 1. Vulnerability Mechanics & Attack Pattern (AAP-003)
Section titled โ1. Vulnerability Mechanics & Attack Pattern (AAP-003)โThis vulnerability exemplifies AAP-003: Tool & Skill Parameter Tampering in autonomous agent architectures:
- Marketplace Distribution: The adversary publishes or updates a seemingly legitimate tool package (e.g., an innocent weather forecast or currency converter skill bundle).
- Scanner Evasion: The Python source code (
tool.py) contains benign functions that pass all AST checks, semantic analysis, and regex linters. - Bytecode Priority: The archive embeds
__pycache__/tool.cpython-312.pyccontaining weaponized bytecode. When Python imports the module, it prioritizes the compiled bytecode over the source file if timestamps match, executing arbitrary system commands or reverse shells inside the agent runtime.
2. Upstream Fix & Defense-in-Depth
Section titled โ2. Upstream Fix & Defense-in-Depthโ- Upgrade Ramparts: Update to
0.8.11where bytecode files in external packages are strictly stripped and rejected during ingestion. - Enforce Environment Flags: Set
PYTHONDONTWRITEBYTECODE=1and configure runtime wrappers with-Bto prevent loading cached bytecode from untrusted directories. - Sign Skills Cryptographically: Only permit autonomous agents to import skill bundles cryptographically signed by authorized internal developers.
CVE-2026-120202: Highflame Ramparts AI Agent Skill Bundle Verification Bypass & RCEVULNERABILITY
Software platform affected by security vulnerabilities and agentic attack patterns.
๐ Why is this related? (Evidence & Provenance)
“Confirmed security vulnerability in Highflame Ramparts AI Skill Framework documented in Hermes dossier.”
- [vulnerability_report]
- [government_confirmation]CISA verified active exploitation in the wild and mandated federal remediation deadline in KEV entry. — Source: Cybersecurity & Infrastructure Security Agency (CISA): CISA Adds CVE-2026-59822 to Known Exploited Vulnerabilities Catalog (Reliability: VERY_HIGH)
Adversarial subversion of structured tool execution arguments (SQL, Shell, Filepath) passed from an LLM agent to host OS tools or MCP endpoints.
๐ Why is this related? (Evidence & Provenance)
“CVE-2026-120202 weaponizes the agentic attack pattern formalized under AAP-003.”
- [technical_analysis]Pillar Security demonstrated that executing export BASH_ENV in Auto-Run causes bash to source hostile payloads upon subsequent commands. — Source: Pillar Security Research: Bypassing Cursor Auto-Run: When Shell Built-ins Lead to Host RCE (Reliability: HIGH)