cu has not published a stable release. Security fixes currently target the latest
maintained main branch and the identified preview release.
| Version | Support |
|---|---|
1.0.0-preview.1 |
Best-effort fixes until superseded |
Unreleased main |
Best-effort fixes |
| Earlier snapshots | Not supported |
Do not report a suspected vulnerability in a public issue, discussion, pull request, or shared capture artifact.
GitHub private vulnerability reporting is the supported reporting channel and a public
launch prerequisite. Before launch, the repository owner must enable it, confirm that
the GitHub API reports it enabled, and make an unauthenticated request to
/mapleluvr/cu/security/advisories/new. The request must redirect to sign-in while
preserving that exact path as its return target. After launch, sign in and use
Security -> Report a vulnerability for all vulnerability details.
If the control is temporarily unavailable, do not send vulnerability details through a public issue, discussion, pull request, or unspecified profile contact. Wait until the maintainer restores the private reporting channel.
Include only the minimum information needed to reproduce the issue:
- affected release or commit;
- Windows, Node.js, and PowerShell versions;
- security impact and required preconditions;
- a minimal reproduction using synthetic content;
- whether observation, native input, archive state, or cleanup was affected;
- sanitized receipts and exit codes.
Do not send real credentials, personal screenshots, raw typed text, private .cu/
archives, helper paths, process identifiers, or window handles. Replace them with
synthetic values before submission.
Reports are especially useful when they demonstrate:
- input emitted without a current actionable observation;
- replay after a partial or indeterminate effect;
- input escaping the requested finite action segment;
- workspace, archive, TEMP, symlink, junction, or run-lock confinement failure;
- exposure of typed text or private helper/target data;
- acceptance of a stale or forged observation, display, lock, witness, or action plan;
- cleanup that deletes state not owned by the current command;
- a claimed completed outcome when native input or cleanup was unproven.
cu is designed to fail closed around stale evidence, unprovable effects, confinement,
and cleanup uncertainty. It is not a prompt-injection detector, authorization service,
malware sandbox, UIAccess boundary, or proof that a fresh action is semantically safe.
Callers remain responsible for deciding what applications and content they are
authorized to control.
Do not run unrestricted full-desktop verification on a personal or ordinary developer desktop. That gate requires an isolated Windows Sandbox or VM containing only synthetic content.
The maintainer will validate scope, coordinate a fix and regression coverage, and agree on disclosure timing through the private report. No response-time SLA or vulnerability bounty is currently offered. Public disclosure should wait until affected users have a reasonable remediation path.