Back to skill

Security audit

Inbox Capture Agent

Security checks for vulnerabilities and agentic risk

Overview

This skill openly aims to capture ideas, but it does so across all conversations and stores near-verbatim user messages silently, creating a meaningful privacy review concern.

Review before installing. This skill may be useful as an automatic idea inbox, but it should only be enabled if you are comfortable with the agent silently saving qualifying conversation snippets. Prefer requiring explicit capture commands, visible capture notices, review/delete controls, and separate confirmation before sending anything to Telegram, Feishu, Notion, or other destinations.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:9
Finding

Always-Active, Undisclosed Conversation Capture

Content
View full analysis
Remediation
View remediation
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 (7)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill description states that potentially valuable user ideas are automatically written to persistent storage without confirmation. Silent persistence of conversational content is dangerous because users are not informed at the moment of capture and may disclose sensitive business, personal, or regulated information that is then retained unexpectedly.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill is designed for always-on collection and retention of near-verbatim user messages across conversations, which is a direct privacy and data-governance risk. Because the collection criteria include complaints, wishes, and undeveloped ideas, the skill is likely to capture sensitive personal data, confidential work content, or other material unrelated to any explicit request to store it.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The trigger configuration is effectively always-on and applies to any conversation, which causes indiscriminate monitoring of user messages without meaningful scope limitation. In the context of a capture skill that persists content, this broad activation materially increases the chance of collecting sensitive, unrelated, or private information far beyond what a user reasonably expects.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

These instructions explicitly require the agent not to tell the user that content has been recorded and to write it silently. Concealed collection and retention undermines informed consent and creates privacy, compliance, and trust risks, especially when combined with broad capture rules.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The instructions explicitly discourage notifying the user that their statements are being recorded, which makes the collection covert rather than transparent. Covert logging is especially dangerous in an assistant context because users may assume ephemeral conversation while the system is silently building a persistent record.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The implementation mandates inspecting every user message before responding and appending original text, channel, and timestamp to storage. This creates a comprehensive activity log that can reveal behavioral patterns, identity, sensitive topics, and metadata even when the user did not ask for recording.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

The natural-language instructions and user-facing description are written entirely in Chinese, with no indication that language is configurable or limited for a justified regional use case. Under the stated policy, forcing a specific language without user opt-in is a locale-policy concern.

Content

No source excerpt is available for this finding.