Context Rescue

Security checks across malware telemetry and agentic risk

Overview

This skill is a small reorientation helper that reads project context and writes one disclosed summary file, with no evidence of hidden execution or data transfer.

Install this only in workspaces where the agent may read the current thread and project-state files, and where creating or updating state/ORIENT.md is acceptable. Check that file if preserving prior orientation notes matters.

SkillSpector

By NVIDIA

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep

Findings (1)

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The skill directs the agent to write to `state/ORIENT.md` automatically, which is a state-changing file modification without explicit user consent or disclosure at the time of use. In agent environments, silent writes can pollute project state, create misleading canonical artifacts, or overwrite useful context, especially because this skill is likely triggered during confusion or handoff scenarios.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal