Skip to content

Back to the catalog

endorlabs.comarticle

What is an agent harness? The software that turns a model into an agent | Blog | Endor Labs

Endor Labs explainer arguing that an AI agent's "harness" (orchestration loop, tool calling/MCP, memory, sandboxes, permission guardrails) - not the underlying LLM - is what determines both capability and attack surface. It walks through concrete harness-level risks (prompt injection as OWASP's LLM01, MCP servers as unvetted dependencies, permission sprawl) and pitches deterministic lifecycle hooks (as used in Claude Code/Cursor via Endor's AURI) as a jailbreak-proof control layer, citing Endor's Agent Security League benchmark where Cursor+Fable 5 beat Claude Code and Codex on both functional and security correctness.

TL;DR Vendor explainer on why the "agent harness" (tools, MCP, sandboxes, hooks) is the real AI-agent attack surface, pitching Endor Labs' AURI hook-based governance product.
Open original ↗

Organisations named here

  • GPT-5.5

Technologies named here

  • Claude Code
  • AURI
  • Cursor
  • LangChain
  • LlamaIndex
  • Codex
  • MCP
  • OpenHands
  • GPT-5.5
  • Hooks
  • Claude Fable 5
  • Git hooks
  • Pre-commit
  • Kubernetes admission webhooks
  • Model Context Protocol
  • Endor Labs
  • Agent Security League
  • OWASP Top 10 for LLM Applications

Techniques named here

  • Prompt Injection
  • Command injection
  • Path traversal
  • SSRF
  • Context rot
  • Malicious npm package injection via agent tool use

Related in AI Security