Security audit

Mulch

Security checks across malware telemetry and agentic risk

Overview

The skill openly sets up a local project memory workflow; users should ensure saved records do not contain secrets or private data.

Install only if you want agents to maintain project memory. Before using it, set a team rule that Mulch records must be short, validated, and free of secrets, credentials, personal data, private customer details, or raw proprietary snippets that should not be shared through git.

SkillSpector

By NVIDIA

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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

Findings (2)

Vague Triggers

Medium

Confidence: 83% confidence
Finding: The skill instructs agents to record learnings under broad triggers such as failures, corrections, missing features, or better approaches without defining clear boundaries on what data may be captured. In practice, this can cause over-collection of sensitive prompts, secrets, internal code details, or user data into persistent memory, especially because the skill encourages use across sessions and teammates.

Vague Triggers

Medium

Confidence: 86% confidence
Finding: The workflow says to record learnings whenever something fails or is learned during work, and again before finishing, but it does not specify decision criteria, approval gates, or data-handling limits. This ambiguity increases the chance an agent will persist transient debugging data, sensitive outputs, or inaccurate conclusions into shared memory, where they may later influence other sessions.

VirusTotal

61/61 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.