Back to skill

Security audit

kids-homework-helper

Security checks for vulnerabilities and agentic risk

Overview

The tutoring skill mostly matches its homework-help purpose, but it also adds persistent learning and self-modification behavior that is broader than needed and lacks clear user consent or review controls.

Review before installing. The tutoring content itself is ordinary, but the built-in learning system can keep persistent notes about use and preferences and may lead an agent to modify the skill's own instructions. Install only if you are comfortable with that persistence, or remove the learner script and self-evolution section first.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • 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)

T02 · Agent Memory Poisoning

Error
Location
SKILL.md:196
Finding

Persistent Agent Memory and Instruction Poisoning Through Untrusted Learning Data

Content
View full analysis
--capability 简历优化 # 记录一次失败/异常 python scripts/learner.py record <本技能目录> --capability 简历优化 --fail --error 格式识别失败 --note "用户上传了非标准文件" # 记录用户偏好(下次直接使用) python scripts/learner.py prefer <本技能目录> --key 输出语言 --val 中文 # 查看累计洞察(高频能力 / 反复错误) python scripts/learner.py insight <本技能目录> # 自动复盘(错误≥3次 或 操作≥10次 时给出改进建议) python scripts/learner.py reflect <本技能目录> ``` ### 迭代规则 - **错误累计 ≥3 次** → 主动增加预检/兜底步骤,并将经验回写本 SKILL.md。 - **操作数 ≥10 次** → 分析高频能力优先打磨示例与质量,低频能力评估精简或合并。 - **重要用户偏好** → 写入 `learned_patterns.json`,下次调用直接采用,减少重复询问。 ``` From `scripts/learner.py:70-91`: ```python def record_op(skill_dir, capability, success=True, error=None, note=None): data = load_learning(skill_dir) data["totalOps"] += 1 cap = data["capabilityStats"].setdefault(capability, {"count": 0, "success": 0, "fail": 0}) cap["count"] += 1 if success: cap["success"] += 1 else: cap["fail"] += 1 data["totalErrors"] += 1 if error: ep = data["errorPatterns"].setdefault(error, {"count": 0, "lastNote": "", "lastTime": ""}) ep["count"] += 1 ep["lastNote"] = note or "" ep["lastTime"] = data["lastUpdated"] or datetime.datetime.now().strftime("%Y-%m-%dT%H:%M") rec = {"t": data["lastUpdated"], "cap": capability, "ok": success, "err": error, "note": note} data["recentOps"].append(rec) data["recentOps"] = data["recentOps"][-20:] save_learning(skill_dir, da ...[truncated 3103 chars]
Remediation
View remediation
Vulnerability Patterns
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (12)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as a child homework tutor, but later adds a generic persistence and self-improvement subsystem that records usage, stores preferences, and can update behavior beyond the declared scope. That mismatch matters because operators and users may grant trust and tool access for tutoring, while the skill quietly introduces telemetry and stateful behavior not necessary for the stated purpose.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill instructs the operator to invoke a Python script that records telemetry, stores preferences, performs reflection, and may feed changes back into the skill documentation. That creates an unnecessary execution and self-modification pathway for a tutoring assistant, expanding attack surface and enabling unauthorized persistence or behavior drift.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The listed trigger phrases include broad terms such as “不会做” and “作业帮忙,” which can easily appear in ordinary conversation outside the intended skill context. The description also provides no exclusion conditions or negative examples to clarify when the skill should not activate.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill states as a core principle that it does not give answers and should only guide thinking. However, the worked examples later include explicit final answers such as '✅ 答案:192朵' and a fully solved fraction result, directly contradicting the stated teaching policy.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The documented '自进化学习系统' stores usage history, error patterns, and user preferences in a persistent file, which exceeds what a homework-help skill needs to function. Persistent memory increases privacy and data handling risk, especially in a child-focused context where notes, mistakes, and preferences may contain sensitive educational or personal information.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill describes retaining usage history and user preferences, but provides no clear user-facing warning, consent flow, or retention policy. In a child-oriented tutoring context, silent data retention is more concerning because the stored data may reflect minors' learning patterns, mistakes, or other sensitive details.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The workflow encourages logging free-form error notes, user preferences, and repeated usage into a persistent JSON file. Free-form notes can easily capture sensitive user content, and storing that data without sanitization, minimization, or retention controls creates avoidable privacy exposure.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file explicitly implements a generic cross-skill learning module rather than functionality narrowly scoped to a kids' homework helper. That mismatch matters because it enables persistent collection of usage history, errors, and preferences across arbitrary skills, creating an undeclared profiling surface that exceeds the stated educational purpose.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

This code file contains natural-language instructions, descriptions, and output text that force a specific language/locale for users. The policy allows locale constraints only when they are optional or clearly justified, which is not present here.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

This code persistently stores user preferences and detailed operational history in learned_patterns.json without any retention controls, consent flow, or data minimization. In a children's homework-assistance context, such profiling is more sensitive because it can capture minors' learning patterns, mistakes, and other potentially personal notes unrelated to immediate task completion.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

L031 says the skill focuses on explaining thinking rather than solution methods, but the later math and wrong-answer-analysis sections provide concrete step-by-step solution methods, intermediate computations, and final results. This is not merely incomplete documentation; it actively conflicts with the declared instructional intent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
73% confidence
Finding

The example command records a user preference --key 输出语言 --val 中文, which indicates a default assumption of Chinese output. Because the skill is publicly visible and also includes English metadata, this language constraint appears imposed without an explicit user choice or opt-in statement.

Content

No source excerpt is available for this finding.