Security audit

部署故障分析及解决助手

Security checks across malware telemetry and agentic risk

Overview

The skill matches its troubleshooting purpose, but it automatically preserves and republishes potentially sensitive deployment logs without enough user control or retention limits.

Install only in an environment where deployment logs may be stored locally and added to operational knowledge bases. Redact secrets, hostnames, customer data, IPMI/root passwords, tokens, and personal identifiers before use, and review or delete the raw archive, generated reports, Excel records, and MySQL inserts as needed.

SkillSpector

By NVIDIA

Vulnerability Patterns

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

Findings (6)

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The skill instructs unconditional archiving of all user-provided files and text into a persistent local directory before any analysis. Troubleshooting logs often contain credentials, internal hostnames, IPs, tokens, and customer data, so automatic retention without consent, minimization, or retention limits creates a real privacy and data-exposure risk.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The skill says new fault data should be written into Excel and optionally synchronized into MySQL knowledge bases, but it provides no general warning that user-submitted content may become persistent operational records. Because incident logs can embed sensitive environment details and personal identifiers, this creates unauthorized retention and secondary-use risk.

Missing User Warnings

Medium

Confidence: 82% confidence
Finding: The script writes Word, JSON, and Excel files directly to caller-supplied paths without any confirmation, destination restrictions, or user-facing disclosure of all artifacts created. In an agent setting, this can lead to unintended persistence of sensitive log content or overwriting files in accessible locations, especially because fault reports may include error messages, task IDs, timestamps, and root-cause details derived from deployment logs.

Ssd 3

Medium

Confidence: 97% confidence
Finding: The workflow mandates indiscriminate local preservation of all raw inputs in a long-lived directory, which materially increases the chance of sensitive log leakage through later access, backup propagation, or misuse by other tools. Since the content is natural-language driven, a user may unknowingly cause secrets or regulated data to be stored permanently.

Ssd 3

Medium

Confidence: 92% confidence
Finding: The skill requires returning a fault summary and sending a generated Word report in the same reply, which can cause direct exfiltration of sensitive log contents back through the chat channel. In this context the report is built from raw diagnostic material, so without sanitization it may re-expose secrets, infrastructure topology, internal paths, and personal data.

Ssd 3

Medium

Confidence: 93% confidence
Finding: The workflow explicitly tells the agent to display a complete ingestion draft containing full error text, deployer identity, resource-pool names, version, and branch metadata. This aggregates sensitive operational and personal information into one response, increasing exposure risk to anyone with access to the conversation.

VirusTotal

62/62 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.