You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I noticed that ARTEX's current target gate does not fully enforce the authorized testing scope.
As documented in intercept/target_extract.go, the gate identifies targets declared in tool inputs, but it cannot prevent a process from connecting to an undeclared destination. This is particularly relevant when agents execute shell commands or launch external scanning tools.
The issue is made more significant by the fact that the target gate is disabled by default for new tasks.
This creates a gap between the scope an operator intends to authorize and the network activity an autonomous agent can actually perform.
I'd suggest strengthening this with network-level restrictions so that all agent processes, including external tools and subprocesses, are subject to the same authorized target boundaries.
Expected outcome: Agents should be unable to communicate with destinations outside their explicitly authorized scope, even when a connection bypasses ARTEX's application-level target checks.
Feature request: enforce network-level egress restrictions so that every agent process — including shell commands, subprocesses and external scanning tools — can only reach destinations explicitly declared in the task's authorized scope, closing the gap where a connection bypasses the application-level target gate described in intercept/target_extract.go (a gap widened by that gate being disabled by default for new tasks). The requested outcome is that agents cannot communicate outside their authorized scope even when application-level checks are skipped. I'm not certain how the existing egress/ and guard/ packages relate to this today, so a short note on how this would layer with them would help scope the work.
🤖 automated reply from the repo bot. A human maintainer reads everything. Ping me in any comment with @artex-bot.
Activity
github-actions commentedon Nov 20, 2026
Feature request: enforce network-level egress restrictions so that every agent process — including shell commands, subprocesses and external scanning tools — can only reach destinations explicitly declared in the task's authorized scope, closing the gap where a connection bypasses the application-level target gate described in intercept/target_extract.go (a gap widened by that gate being disabled by default for new tasks). The requested outcome is that agents cannot communicate outside their authorized scope even when application-level checks are skipped. I'm not certain how the existing egress/ and guard/ packages relate to this today, so a short note on how this would layer with them would help scope the work.
🤖 automated reply from the repo bot. A human maintainer reads everything. Ping me in any comment with
@artex-bot.