Security audit

WorkRally

Security checks across malware telemetry and agentic risk

Overview

This skill is a disclosed WorkRally CLI guide for AI media creation and workspace management, with expected but sensitive project, asset, credential, and canvas actions.

Install only if you trust WorkRally and intend to give an agent access to your WorkRally account. Use least-privilege API keys where possible, keep the config file private, and require explicit confirmation before deletes, permanent recycle-bin actions, uploads/downloads of sensitive media, or canvas overwrite/clear operations.

SkillSpector

By NVIDIA

Vulnerability Patterns

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

Findings (5)

Context-Inappropriate Capability

Medium

Confidence: 95% confidence
Finding: The document explicitly advertises a generic passthrough that can list, describe, and invoke any MCP tool, which materially expands capability beyond the stated WorkRally-focused scope. In an agent setting, this creates a tool-confusion and scope-bypass risk: an LLM following the skill may call unintended tools with sensitive or destructive side effects, especially because the examples normalize unrestricted arbitrary invocation.

Missing User Warnings

Medium

Confidence: 89% confidence
Finding: The skill exposes commands for upload, download, soft deletion, and overwrite-style canvas mutations, including operations that immediately affect shared collaborative state. Without prominent user-facing warnings or confirmation guidance, an agent could perform destructive or privacy-impacting actions that alter user data, shared workspaces, or retrieve assets unintentionally.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The skill instructs users to persist API keys to a local config file and references the exact storage path, but does not prominently warn about credential persistence, file permissions, or multi-tenant/container risks. In agent or shared environments, this can lead to accidental credential exposure through filesystem access, logs, reused volumes, or downstream tools reading the config.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The guide documents destructive `canvas build-draft` operations in a real-time collaborative environment, but the warnings are easy to miss relative to the command examples. An agent following these examples could delete or overwrite other users' live canvas content without explicit confirmation or safer defaults, causing integrity loss in shared workspaces.

Missing User Warnings

Medium

Confidence: 83% confidence
Finding: The example `canvas build-draft --nodes '[]' --mode overwrite` is a destructive operation that appears in a quick-reference table without an adjacent warning that it deletes all canvas content. In agent-driven execution, terse cookbook examples can be copied verbatim, increasing the chance of accidental data loss or irreversible overwrite when a user merely asks to modify or reset part of a canvas.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.