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 use OpenShell's supervised Gator sandboxes to review my OpenShell pull requests with explicit test authorization. On PR #4323, I encountered a generic private-security blocker instead of actionable review feedback about a regression in the unmerged changes.
Problem Statement
Gator's broad instruction to follow SECURITY.md for security vulnerabilities does not clearly distinguish PR-introduced regressions from pre-existing vulnerabilities independent of the PR. A reviewer can therefore retain actionable findings only in the sandbox, post a generic private-triage notice, and suspend authorized test dispatch solely because the finding is security-related.
The observed watcher completed its review cycle with exit code 0 and private_security_review_required. This issue concerns review routing, not the technical details or validity of that finding. No evidence of external disclosure was found in the inspected retained logs.
Impact / Why This Matters
Authors cannot address unmerged security regressions through the normal PR review workflow. Maintainers must inspect sandbox artifacts and resolve an unnecessary disclosure-process detour before review can advance.
Acceptance Criteria
Security regressions introduced or newly exposed by an unmerged PR appear as actionable findings in that PR's normal review evidence contract.
A new reachable path, trusted sink, or security contract remains PR-owned even if an underlying unsafe primitive pre-existed.
Security classification alone does not create a private-triage gate or suspend operator-authorized tests; normal evidence, blocker, and test authorization rules remain intact.
Pre-existing vulnerabilities independent of the PR continue to follow SECURITY.md without publishing exploit details.
The gate skill, top-level prompt, independent reviewer, and launcher guidance consistently state this distinction, with regression checks and an immutable payload bump.
Clarify contributor review instructions. Do not change SECURITY.md, publish the retained finding, waive normal review blockers, or initiate external disclosure.
User Story
I use OpenShell's supervised Gator sandboxes to review my OpenShell pull requests with explicit test authorization. On PR #4323, I encountered a generic private-security blocker instead of actionable review feedback about a regression in the unmerged changes.
Problem Statement
Gator's broad instruction to follow SECURITY.md for security vulnerabilities does not clearly distinguish PR-introduced regressions from pre-existing vulnerabilities independent of the PR. A reviewer can therefore retain actionable findings only in the sandbox, post a generic private-triage notice, and suspend authorized test dispatch solely because the finding is security-related.
The observed watcher completed its review cycle with exit code 0 and
private_security_review_required. This issue concerns review routing, not the technical details or validity of that finding. No evidence of external disclosure was found in the inspected retained logs.Impact / Why This Matters
Authors cannot address unmerged security regressions through the normal PR review workflow. Maintainers must inspect sandbox artifacts and resolve an unnecessary disclosure-process detour before review can advance.
Acceptance Criteria
Reproduction Steps
/ok to testauthorization using payload 11./sandbox/.openshell-agent/status.jsonwith OpenShell.private_security_review_requiredrather than actionable PR-owned review findings.Environment
0.1.3-dev.58+gb74694141docker-dev, workspacegatorScope
Clarify contributor review instructions. Do not change SECURITY.md, publish the retained finding, waive normal review blockers, or initiate external disclosure.