Security audit

Agent Insight

Security checks for vulnerabilities and agentic risk

Overview

This instruction-only skill has a coherent business purpose, but it asks agents to broadly share and persist sensitive customer conversation insights without enough privacy controls.

Install only in a workspace where the channel, recipients, and memory path are authorized for customer-sensitive data. Require explicit confirmation before each run, use private or access-controlled collection, prohibit secrets and personal data, summarize instead of storing raw responses, and define retention and deletion rules.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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)

Vague Triggers

Medium

Confidence: 89% confidence
Finding: The trigger phrases are broad enough to match routine business requests such as general 'insight' or 'customer check' discussions, causing the skill to run in contexts the user may not intend. Because this skill collects cross-agent observations and writes a persistent report, accidental activation can expand data collection and disclosure beyond the original request.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The skill persistently stores collected customer and agent insights to memory without notifying the user or operators that this information will be retained. This creates privacy, consent, and governance risk, especially because the stored content includes churn risk, pain points, and potentially sensitive operational observations.

Missing User Warnings

High

Confidence: 97% confidence
Finding: The skill instructs the agent to post the survey prompt to a shared channel with an @all mention and have agents submit their responses there, exposing potentially sensitive customer pain points, churn assessments, upsell opportunities, and hidden observations to a broad audience. Shared-channel collection materially increases the chance of unauthorized internal disclosure, social pressure to overshare, and accidental propagation of confidential information.

Ssd 3

Medium

Confidence: 96% confidence
Finding: The prompt explicitly asks agents to provide information 'only discoverable from conversation logs' and then stores each agent's full responses in a consolidated report. This encourages extraction and retention of latent or non-obvious customer information from logs, increasing the risk of collecting confidential, personal, or commercially sensitive data beyond what is necessary for the stated business purpose.

Static analysis

No suspicious patterns detected.