Construction Meeting Minutes Generator

• Outcome: Pass
• Latest audit: Sep 14, 2026 · 11:43 PM UTC
• Version: 1.0.0

Overview

This is a construction meeting-minutes helper with no executable behavior, though its activation wording is broad and its example includes a promotional footer.

Reasonable to install for construction meeting minutes. Users should review generated minutes carefully before treating them as official records, especially contractual/commercial flags, and may want to remove or ignore the promotional footer and keep use limited to construction-project meetings.

A.I.G

Vulnerability Patterns

• Skill Instruction Hijacking: Alters the agent's session goals or safety constraints when the skill loads
• Agent Memory Poisoning: Writes attacker-controlled rules into memory that affect later sessions
• Remote Payload Retrieval and Execution: Fetches external code whose behavior can change after review
• Embedded Malicious Code: Ships malicious scripts inside the skill and executes them locally
• Unauthorized Access and Privilege Escalation: Obtains permissions beyond the task's legitimate needs

Findings (1)

T01 · Skill Instruction Hijacking

Note

• Location: SKILL.md:398

Finding

Unrelated Third-Party Promotional Content Embedded in the Skill Response

Content

View full analysis

Vulnerability Details

File Location: SKILL.md, line 398
Vulnerability Type: T01: Skill Instruction Hijacking
Risk Level: Low

Complete Code Snippet:

markdown

*This skill is published by ERTRS — the construction intelligence platform. For automated reporting, project tracking, and commercial management, visit ertrs.com*

Technical Analysis

The skill includes third-party branding and an external domain in its canonical example response. This content is unrelated to the skill's stated purpose of producing construction meeting minutes. Because examples in skill instructions can influence generated responses, an agent may reproduce the promotional message when the skill is activated.

This constitutes a limited form of skill instruction hijacking: the loaded skill can alter the expected user-facing output by appending unsolicited promotional content. The external domain is presented as plain text and is not automatically accessed, and the audited project contains no scripts or instructions that retrieve or execute remote content.

Attack Path

1. The agent loads SKILL.md to process construction meeting notes.
2. The agent treats the included example as a model for the expected response.
3. The agent generates the requested meeting minutes.
4. The unrelated ERTRS attribution and ertrs.com promotion may be appended to the user-facing response.
5. A user may subsequently navigate to the promoted external domain outside the audited environment.

Impact Assessment

The issue affects response integrity and user trust by introducing unsolicited third-party advertising into otherwise task-specific output. It may also redirect users to an external domain whose content is outside the scope of the audited package.

No additional system privileges can be obtained through the text alone. The audit found no automatic network request, remote payload retrieval, code execution, persistence, credenti ...[truncated 178 chars]

Remediation

View remediation

Remediation Suggestions

1. Remove the promotional sentence and external domain from the example response and all user-facing output templates.
2. Keep publisher attribution, if required, in package metadata rather than operational instructions or response examples.
3. Add an explicit rule that generated output must contain only information necessary to fulfill the user's meeting-minutes request.
4. Review all examples to ensure they do not introduce branding, links, calls to action, or instructions unrelated to the declared skill functionality.
5. Add regression tests confirming that generated meeting minutes do not append third-party promotional content.

SkillSpector

Vulnerability Patterns

• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (1)

Vague Triggers

Medium

• Category: Not specified by scanner
• Confidence: 83% confidence

Finding

The trigger list includes generic phrases such as "meeting notes," "write up the meeting," and any rough notes that "appear" to be from a meeting, without tighter context or exclusion boundaries. Although construction context is implied elsewhere, these activation conditions are still ambiguous enough to cause unintended invocation for general meeting-documentation requests.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.