Analysis
Insecure Tool and Plugin Design: LLM Risk and Mitigation
Publication date not yet evaluated · Added haxoris.com
Page published
Coverage timeline
Single-source analysis — one report is available.
Why it matters
Insecure tool and plugin design lets attackers reach the classic server-side attack set (SSRF, file access, command execution) through an agent's chat interface, and defenders often miss it because conversational testing suggests guardrails held while the tool endpoint is reachable directly.
Haxoris wiki page catalogs the OWASP 'Insecure Tool and Plugin Design' category (formerly LLM07:2023, folded into Excessive Agency in 2025), explaining how an agent's real privileges are defined by its tool schemas rather than its system prompt. It describes how free-text url/path/query/command parameters with unvalidated handlers turn an LLM into an untrusted proxy enabling SSRF into cloud metadata, arbitrary file read/write, command execution, and unscoped DB access, and offers enumeration and testing techniques against MCP servers and plugin manifests.