AI Stack Alert High CVE-2026-117573 For AI Eng, AppSec, SOC / IR, CISO AI-assisted
CVE-2026-117573: hagent hagent_arduino/hagent_esp32 MCP tools allow OS command injection
hagent through commit 761eec6 has an OS command injection flaw in the args parameter of its hagent_arduino and hagent_esp32 MCP tools. Crafted args are interpolated into arduino-cli and idf.py commands run with subprocess.run(shell=True). The source states this lets MCP clients run shell commands as the MCP server user; the CVSS base score is 8.5.
Why it matters
Any MCP client, including an AI agent that has been prompt-injected, can run arbitrary shell commands with the privileges of the MCP server process. Hosts running hagent with access to source code, credentials, or build systems are exposed.
Do this
- Inventory hosts running hagent at or before commit 761eec6 and check for a fixed release or commit from the upstream project.
- Until patched, disable or remove the hagent_arduino and hagent_esp32 tools from MCP server configurations, or block agent access to them.
- Replace shell=True and string interpolation with argument-list subprocess calls, and validate args against an allowlist.
- Run the hagent MCP server under a low-privilege account with no access to unrelated secrets or production networks.
- Hunt for unexpected child processes spawned by the hagent MCP server process, especially shell interpreters, and review MCP tool-call logs for unusual args values.
Exploit Clock
Exploitation unlikely (for now)
- EPSS 0.70%
Signals
- NoCISA KEV
- 8.5CVSS
- 0.70%EPSS
- NoPublic PoC
- 34/100Priority
- 70%Confidence
Evidence
“Attackers or prompt-injected agents can supply crafted args interpolated into arduino-cli and idf.py commands run with subprocess.run(shell=True) to execute commands as the MCP server user.”
“allows MCP clients to execute shell commands by injecting metacharacters into the args parameter of the hagent_arduino and hagent_esp32 tools”
Sources
Affected / related
hagent, hagent_arduino, hagent_esp32, arduino-cli, idf.py, OS command injection
Published 2026-10-27 14:16 UTC · updated 2026-11-08 02:41 UTC · AI-assisted summary, checked against quoted sources · How we score