Replies: 2 comments
|
This is the #1863 family again, and the framing matters: approval gates and workspace-write limits are consent/routing UX, not OS security boundaries. Delegating execution to a user-privileged external shell (explorer/start/open) sidesteps both ? the agent isn't running the command, the user's own context is. Fix directions, in order of leverage:
Happy to help spec a "shell-launcher risk" check for dsh-plugin-doctor's preflight (detect |
|
Toolized: dsh-plugin-doctor v1.11.0 adds a npx dsh-plugin-doctor check ./my-plugin --json
# ? shell-launcher: warn | src/index.ts (explorer.exe) ?31/31 tests, doctor self-check exit 0 (its own cmd.exe wrapper shows as an honest WARN). Release: https://github.com/zoahdev/dsh-plugin-doctor/releases/tag/v1.11.0 ? the OS-level sandbox direction (#1789 restricted token) remains the real cure; this check makes the risky channel visible before shipping. |
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
同类问题已有讨论,此贴关闭
All reactions