Back to skill

Security audit

meta-self-evolution-metrics

Security checks for vulnerabilities and agentic risk

Overview

This skill is not clearly malicious, but it asks the agent to persist user preferences and potentially rewrite its own SKILL.md without enough scoping or user control.

Install only if you are comfortable with a skill that keeps local cross-session learning state and may propose or perform changes to its own instructions. Before use, require explicit confirmation before saving preferences or notes, treat learned_patterns.json as untrusted data, and do not allow automatic SKILL.md rewrites without human review.

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

Warning
Location
scripts/learner.py:5
Finding

Unvalidated Persistent Notes Can Poison Future Agent State

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。 ``` ### Technical Analysis The `record()` function appends the supplied `note` directly to a persistent JSON file without validating its type, length, origin, or semantic content. There is no distinction between inert historical data and text that could be interpreted as instructions by an agent. ...[truncated 3174 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (12)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill claims broad self-validation, self-reflection, super-agent orchestration, and continuous self-evolution capabilities, but the documented concrete behavior is mostly local logging and persistence. This mismatch is dangerous because users and orchestrators may grant elevated trust or autonomy based on claimed safeguards that are not actually implemented, while hidden persistence behavior is not transparently disclosed.

Content

No source excerpt is available for this finding.

Self-Modification

High
Category
Rogue Agent
Confidence
90% confidence
Finding

The recommendation path includes selfpatch, indicating self-modification or automatic expansion behavior. In a skill context, self-modification is particularly dangerous because it can change operational logic or instructions over time, potentially bypassing prior review and increasing the blast radius of prompt injection or configuration mistakes.

Content

Scanner excerpt · SKILL.md (reported line 20)May include surrounding context.

md
- 显性工作流步骤(3 步):
  1. **加载(load)**:读取 meta-evolver 的 evolver_patterns.json(封闭/已知缺口/历史动作)。
  2. **度量(evaluate)**:算覆盖度、停滞、漂移,给出推荐动作。
  3. **建议(recommend)**:饱和→selfpatch 扩展矩阵;停滞→聚焦 Top 缺口;健康→继续。

## 增强点(超越教师)
1. **可靠自验证**:每步产出后用 `reason-verify` 做命题一致性/事实锚定校验,reliability<0.8 即回退重做。

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
94% confidence
Finding

The skill documents file-writing behavior via persistent learning state (learned_patterns.json) and even suggests modifying SKILL.md, but the manifest declares no permissions or allowed-tools. This creates a trust and review gap: operators cannot accurately constrain or approve the skill’s write capabilities, increasing the risk of unexpected persistence or file modification.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill is framed as a metrics/evaluation helper, but the documentation expands its role into persistent learning and changing its own documentation. This scope drift matters because a seemingly analytical skill is also positioned to retain state and alter future behavior, which changes its risk profile and can bypass user expectations or deployment controls.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill describes automatic persistence of user preferences and usage history but does not provide a clear privacy warning, consent flow, or explanation of retention and reuse. This is dangerous because users may unknowingly provide personal or operational data that is then stored across sessions and reused without transparency.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill persistently stores user preferences and usage history even though its stated purpose is metrics/self-evolution analysis. Retaining cross-session user data without strong necessity increases privacy exposure and creates a durable record that could leak sensitive workflow details or user-specific preferences.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill plainly instructs persistent storage and reuse of user preferences across sessions, which creates data-retention and leakage risk. In this context, the danger is amplified because the retained information can influence future agent behavior while remaining outside the user’s immediate visibility, increasing privacy and integrity concerns.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The instructions encourage recording user preferences for future automatic reuse, yet do not warn that this creates persistent state or discuss downstream effects. That omission increases the chance of unauthorized profiling, stale preference reuse, or accidental storage of sensitive information.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The rule to rewrite SKILL.md gives the skill a self-modifying documentation pathway that extends beyond a metrics role. Self-modification is risky because it can silently expand instructions, normalize unsafe behavior over time, or make later review difficult by mutating the skill’s own specification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The entire skill file is written in Chinese, including headings and operational descriptions, with no indication that language selection is optional or that the skill is intended only for a Chinese-speaking or region-specific context. This creates a natural-language policy concern because it imposes a locale/language on users without documented opt-in or justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

The module docstring is written only in Chinese and presents the skill identity and purpose in that locale without offering any language choice or documenting a justified region-specific constraint. This can violate language/locale policy when users are not explicitly opting into Chinese-only behavior.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

This code persists operational metadata and free-form notes to a local JSON file, but there is no confirmation prompt, print/log disclosure before the write, or inline warning that user-provided notes will be stored. Because the file may contain invocation notes and failure history, the silent write is a safety-relevant data-retention behavior in a code file.

Content

No source excerpt is available for this finding.