Sapling SCM EUVD-2026-91648

Exploitation requires the victim to clone a maliciously crafted Git repository whose Git subtree URL contains embedded control characters, and then to run an otherwise read-only Sapling command (sl log, sl blame, or sl annotate) against it. Limiting factors: the vector is local (AV:L) and depends on user interaction (UI:R) - the attacker cannot reach the victim remotely and must socially engineer the clone of an untrusted repository; it affects only Sapling SCM versions before v0.2.20260929-102736, so patched installations are not vulnerable; and exploitation is not automatable (SSVC Automatable=no) because each target must individually obtain and operate on the hostile repo.

This is a genuine high-impact code-execution flaw but NOT a drop-everything emergency, and the signals are internally consistent rather than conflicting. The CVSS 3.1 vector (AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H, 7.8 HIGH) correctly models the real attack: the vector is LOCAL (AV:L) and gated on user interaction (UI:R) because a victim must first clone and then operate on a maliciously crafted repository - it is not a remotely reachable network service bug. CVSS PR:N indicates no authentication is needed on the victim's side, which is accurate since any developer running sl log/blame/annotate is enough. SSVC reinforces a measured priority: Exploitation=none (no observed in-the-wild activity), Automatable=no (each victim must be socially engineered into cloning a specific hostile repo), but Technical Impact=total (full code execution). No EPSS score and no KEV listing were provided, and SSVC Exploitation=none aligns with that absence. There is no POC indicator in the input. Net: a real priority for organizations that actually use Sapling SCM and pull untrusted repositories, but the local vector, required user interaction, and non-automatable nature keep it well below worm-grade/network-RCE urgency.

An attacker publishes a malicious Git repository containing a subtree URL with embedded control characters and lures a developer into cloning it (e.g., via a pull request, a shared fork, or a seemingly benign open-source project). When the victim later runs an ordinarily safe read-only command such as sl log, sl blame, or sl annotate, the control sequence is interpreted and attacker-chosen code executes in the developer's context. No public exploit was identified in the provided data, and exploitation requires the local user interaction reflected in the CVSS UI:R/AV:L metrics.