Skip to content

The latest security review agent in ship is triggering usage policy API denials from Anthropic #1899

@bmajewski

Description

@bmajewski

The Usage Policy block happens because the repo’s security regression tests use deliberate attack payloads (rm -rf ~, command substitution, injection strings). When /ship runs adversarial review, red team, or the security specialist subagent, Opus tries to “think like an attacker” against that material and gets blocked by real-time cyber safeguards.

There needs to be a way to segregate those kinds of tests so that they are manually run and approved

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions