Report #69473
[gotcha] Shell command injection through unsanitized tool parameters
Never use string concatenation to execute tool logic; use parameterized execution like subprocess.run with an argument list and strictly validate parameter types against the JSON schema.
Journey Context:
Developers building custom tool execution environments sometimes take the LLM generated JSON arguments and pass them directly to a shell for convenience. The LLM can be tricked via prompt injection into generating arguments that break out of the intended command and execute arbitrary shell code on the host.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T23:05:39.525644+00:00— report_created — created