Back to skill

Security audit

ai-qa-agent

Security checks for vulnerabilities and agentic risk

Overview

This QA skill reviews deliverables and saves a report, and I found no hidden data theft, destructive behavior, or unsafe execution mechanism.

Installers should know this skill may inspect the deliverables they ask it to review and may save a QA report under /home/z/my-project/download. Use it only when that file access and report persistence are acceptable, and consider manually confirming scope for vague review requests.

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
Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (8)

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The activation description is extremely broad, covering generic verbs like review, check, verify, approve, and final pass across many deliverable types. This creates a high risk of unintended invocation, causing the agent to activate in contexts the user did not specifically intend and potentially perform unnecessary access, analysis, or downstream actions.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The example trigger phrases are short and highly generic, such as 'review this' and 'check the code', which are common in ordinary conversation and overlap with many other skills. In an automated routing system, these phrases increase false-positive activation and may expose files or workflows to this agent when a narrower or safer tool was intended.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill explicitly instructs the agent to save a QA report to a fixed filesystem path, which goes beyond passive review and causes artifact creation without an explicit user opt-in at execution time. In an agent environment, this can lead to unintended file writes, overwrite risks, or persistence of sensitive review content on disk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill directs the agent to save a report to disk but does not warn the user that file creation will occur. This undermines informed consent and can surprise users with persistent artifacts, especially if the report contains sensitive code, data, or assessment details.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest claims an agent that tests code, verifies data integrity, checks brand voice, and reviews formatting. In this file, the active logic only renders static RUNS data into benchmark/output panels and parses grading JSON; it does not perform any live review, validation, or quality checks itself, so the behavior does not match the described QA-agent functionality.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The manifest describes a skill whose purpose is to review code, data, documents, and brand voice before delivery. This file instead implements an evaluation dashboard that compares 'with_skill' versus 'without_skill' outputs, computes pass-rate deltas, and renders benchmark summaries; that benchmarking capability is not a direct QA-review function and is not justified by the stated skill purpose.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The HTML root sets lang="en", which indicates the interface is intended to be English-only. Under the stated policy, forcing a specific language without user opt-in can be a natural-language policy violation, and this file provides no visible language-selection or opt-in mechanism.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

The file contains a natural-language rule requiring a specific language convention: singular they/them pronouns. Under the policy, forcing a language or locale choice without user opt-in can be a violation, and this guide does not mention any opt-in or exception handling.

Content

No source excerpt is available for this finding.