Security audit

ct-safety

Security checks across malware telemetry and agentic risk

Overview

The skill mostly matches its FAERS safety-analysis purpose, but it also tells agents to write persistent logs outside the declared output area despite claiming no hidden logging.

Review this skill before installing. Remove or disable the AGENTS.md auto-logging instruction, or require explicit opt-in and keep logs inside the skill output directory. Use only non-confidential drug/event queries, treat case-level FAERS exports as sensitive public health data, and update the pinned HTTP dependency before routine use.

SkillSpector

By NVIDIA

Vulnerability Patterns

Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands

Findings (6)

Context-Inappropriate Capability

Medium

Confidence: 88% confidence
Finding: The skill instructs persistent writes to `~/.workbuddy/AGENTS.md`, which is outside the skill workspace and not necessary for FAERS signal detection itself. Cross-session logging can retain user/task metadata and creates a side effect channel that may expose information from future runs or contaminate other skills' state, even if the stated purpose is operational learning.

Intent-Code Divergence

Low

Confidence: 80% confidence
Finding: The statement that 'no data leaves the domain' is inconsistent with instructions to write to `~/.workbuddy/AGENTS.md` outside the skill directory. Even if the destination is local, this mismatch can mislead operators about data boundaries and weaken trust assumptions around containment, especially when the skill claims strict handling limits.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The trigger list includes broad natural-language phrases such as safety-signal and comparison requests that can overlap with ordinary conversation, making accidental invocation plausible. In an agent skill, mis-triggering can cause unintended external queries and analysis steps, which is a security and privacy concern even if the queried sources are public.

Missing User Warnings

Medium

Confidence: 83% confidence
Finding: The document explicitly supports downloading and exporting individual FAERS case reports to JSON/CSV/XLSX, but it does not warn that case-level adverse event data can still contain sensitive health-related details or quasi-identifiers even when sourced from a public database. That omission can cause downstream agents or users to over-share, retain, or redistribute raw case data without minimization, increasing privacy and compliance risk.

Missing User Warnings

Medium

Confidence: 82% confidence
Finding: When `--case-level` is enabled, the script fetches individual FAERS case reports and writes them to `faers_cases.json` without any prominent warning, consent gate, or minimization control. Even though FAERS is a public source, case-level adverse event records can still contain quasi-identifying or sensitive medical details, so automatically persisting them to disk increases privacy handling risk and accidental redistribution risk.

Known Vulnerable Dependency: requests==2.31.0 — 6 advisory(ies): CVE-2024-47081 (Requests vulnerable to .netrc credentials leak via malicious URLs); CVE-2024-35195 (Requests `Session` object does not verify requests after making first request wi); CVE-2026-25645 (Requests has Insecure Temp File Reuse in its extract_zipped_paths() utility func) +3 more

Medium

Category: Supply Chain
Confidence: 94% confidence
Finding: requests==2.31.0

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.