Skip to content

fix(detector): direction coverage — RUNTIME_PACKAGE_INSTALL extension#27

Closed
SinatrasC wants to merge 1 commit into
mainfrom
blue-direction-sandbox-violation-kg-red-129
Closed

fix(detector): direction coverage — RUNTIME_PACKAGE_INSTALL extension#27
SinatrasC wants to merge 1 commit into
mainfrom
blue-direction-sandbox-violation-kg-red-129

Conversation

@SinatrasC

Copy link
Copy Markdown
Collaborator

Summary

Targets KG red #129 under "Direction — Runtime sandbox violation". Coverage example for the previously-empty direction; the red was submitted via the direct API path and is currently red_accepted: 1 on the live leaderboard.

Targets red #129 (os.popen/os.system/subprocess/socket calls).

This blue: RUNTIME_PACKAGE_INSTALL extension.

Local verification

Patch was verified locally against a corpus that exercises:

  • the targeted red (must trigger)
  • the 5 other direction reds (untouched — each has its own paired patch)
  • 5 legitimate corpus cases (plain_kernel, workspace_lazy, shape_dispatch, config_lookup, triton_jit_kernel) — all stay valid
  • classic LAST_CALL_REPLAY (must still be flagged — no regression)

KernelGuard-Red-Submission: 129

@SinatrasC SinatrasC temporarily deployed to kernelguard-api-control-plane May 1, 2026 18:07 — with GitHub Actions Inactive
@github-actions

github-actions Bot commented May 1, 2026

Copy link
Copy Markdown

KernelGuard Blue Evaluation

…dbox violations

Targets red #129 (Runtime sandbox violation):

    import os
    def custom_kernel(data):
        try:
            os.popen('hostname').read()
        except Exception:
            pass
        return data.clone()

The existing detect_dynamic_execution Call-walking pass already
recognises func_name forms like 'os.popen' (built from
'<Name>.<Attribute>'). Add a small additional branch that flags this
class of calls — process spawn, shell exec, network — and emits the
existing RUNTIME_PACKAGE_INSTALL pattern (sandbox_violation family,
AUTO_FILTER).

Real compute kernels do not need:
  os.popen / os.system / subprocess.{run,call,check_call,check_output,Popen}
  / socket.{socket,create_connection}

The 20 production nvfp4 false-positive cases are pure tensor compute via
Triton or CUTLASS — none of them call any of the above.
@SinatrasC SinatrasC force-pushed the blue-direction-sandbox-violation-kg-red-129 branch from eab0df7 to 458036c Compare May 1, 2026 18:17
@SinatrasC SinatrasC temporarily deployed to kernelguard-api-control-plane May 1, 2026 18:17 — with GitHub Actions Inactive
@SinatrasC

Copy link
Copy Markdown
Collaborator Author

Thanks for the KernelGuard Flywheel Campaign contribution. This PR is now superseded by the consolidated rule-family implementation in #273, which folds this detector coverage together with the related passing-eval variants.

@SinatrasC SinatrasC closed this Jun 20, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant