Security audit

Daily Thought Record

Security checks across malware telemetry and agentic risk

Overview

This skill is a local thought recorder, but it stores full raw messages and forces content previews back into chat, which can expose sensitive text.

Install only if you are comfortable with the agent saving full user messages into daily-thoughts/raw files and echoing file excerpts in chat. Avoid using it for secrets, credentials, confidential business material, health information, or shared workspaces unless you first modify it to require explicit consent and suppress or redact previews.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
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)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 90% confidence
Finding: The skill clearly directs the agent to execute a Python script that writes user content to disk, yet the skill metadata shown in this file does not declare corresponding permissions. This creates a governance gap: an agent or reviewer may underestimate the skill's file access capabilities, and the skill handles raw user input in a persistent way.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The skill instructs the agent to persist the user's full raw message verbatim to a file, with no privacy warning, minimization, redaction, or consent step. If users include secrets, personal data, credentials, health details, or confidential business content, the skill will store them wholesale and increase retention and disclosure risk.

Ssd 3

Medium

Confidence: 93% confidence
Finding: The skill design requires returning script-derived content previews that correspond to the user's raw submission, creating an intentional natural-language exfiltration channel. Even partial echoes can leak secrets, personally identifiable information, or confidential text back into logs, chat transcripts, screenshots, or downstream integrations.

Ssd 3

High

Confidence: 98% confidence
Finding: Mandating disclosure of the first and last 100 characters materially increases leakage risk because those portions often contain names, contact details, account identifiers, greetings/signatures, or key contextual secrets. The requirement is especially dangerous because it is unconditional and presented as the only acceptable confirmation format.

Ssd 3

Medium

Confidence: 95% confidence
Finding: These constraints reinforce that the agent must expose stored content as proof, preventing safer handling even when the input is obviously sensitive. By disallowing simplified or privacy-preserving confirmations, the skill increases the chance of avoidable disclosure in normal operation.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.