Agent Harnesses Need Security Rule: Untrusted Input ≠ Privileged Access

MoreIndependent5967 · reddit · 2026-08-28

This post proposes a core security principle for Agent harnesses: untrusted external content should never have side-effecting capabilities.

The Risk

Currently, main agents often have privileged access (filesystem, shell, Git, APIs) while also browsing arbitrary web or reading external emails, creating a dangerous trust model.

Proposed Architecture

External content should be processed by read-only subagents:

These subagents pass structured results to the Main Agent, which handles reasoning and delegates to privileged executors.

Key Rules

These restrictions must be enforced by the runtime, not just by prompting.

Original post →

More from coding & agent

coding & agent channel →