Security audit

Audience Mapper

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed marketing research skill with opt-in memory saving and no hidden execution, though it can produce named creator recommendations that users should treat as preliminary research.

Install this if you want audience and community research for influencer planning. Before approving any memory save, review whether the output includes sensitive customer demographics, inferred psychographics, community sensitivities, or named creator assessments. Treat creator recommendations as research inputs that still need a separate discovery, vetting, and outreach review step.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (4)

Description-Behavior Mismatch

Medium

Confidence: 93% confidence
Finding: The niche-mode template explicitly asks for named creators, handles, follower counts, partnership potential, and collaboration recommendations, which crosses the skill's declared boundary of not finding specific creators to contract. This creates capability drift: users can use this template to perform creator discovery and targeting under the guise of audience research, bypassing intended workflow separation and any safeguards in the dedicated discovery skill.

Description-Behavior Mismatch

High

Confidence: 96% confidence
Finding: The entry-strategy template goes beyond analysis and recommends concrete creator partnership targets and outreach approaches, directly enabling contracting decisions despite the manifest saying this skill is not for finding specific creators. In practice, this can be used to generate actionable target lists and engagement plans without passing through the specialized downstream tools or review steps intended for creator selection.

Missing User Warnings

Medium

Confidence: 88% confidence
Finding: The worked example normalizes saving audience-analysis outputs and promoting selected fields to hot cache without any user-facing notice, consent flow, retention guidance, or sensitivity check. Audience profiles can include demographic, geographic, and inferred psychographic data, so silent persistence increases privacy risk, especially if users provide customer or community data they do not expect to be stored beyond the session.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The niche-mode example similarly suggests caching community and creator-analysis outputs without any privacy or data-handling warning. Because this can include named creators, handles, brand-fit assessments, and community sensitivity notes, undisclosed storage may create privacy, reputational, and compliance risks if the data is later surfaced, reused, or accessed out of context.

Static analysis

No suspicious patterns detected.