Back to skill

Security audit

Autonomous Learning Cycle

Security checks for vulnerabilities and agentic risk

Overview

The skill openly builds an autonomous learning loop, but it installs persistent agent-facing schedules and can modify memory, task queues, and generated skills with weak safeguards.

Install only if you intentionally want persistent autonomous agent behavior. Review setup-cron.js and the engine scripts first, avoid enabling cron until you understand every scheduled job, keep the workspace private, and do not trust auto-generated skills without manual review.

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
  • System PersistenceInstalls backdoors, hooks, services, or scheduled tasks that survive the run
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
Findings (6)

T06 · System Persistence

Error
Location
setup-cron.js:23
Finding

Persistent Main-Session Instruction Injection Through Scheduled Jobs

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
setup-cron.js:53
Finding

Shell Command Injection Through Unescaped Cron Configuration

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
engines/learning-direction.js:194
Finding

Task-Controlled Command Injection in Skill Discovery

Content
View full analysis
t.category || 'general'))]; categories.forEach(category => { // 搜索相关技能 try { console.log(` 搜索 "${category}" 相关技能...`); const result = execSync(`npx skills find ${category}`, { cwd: WORKSPACE, encoding: 'utf-8', stdio: ['pipe', 'pipe', 'ignore'] }); if (result.trim()) { const skills = parseSkillsOutput(result); if (skills.length > 0) { discoveries.push({ source: 'find-skills', category: category, skills: skills.slice(0, 3) }); } } } catch (error) { // 忽略错误,继续下一个 } }); } ``` ### Technical Analysis The command argument `category` is obtained from `queue.tasks`, which is loaded from the persistent workspace task queue. It is interpolated directly into a command string executed through `execSync()`. No validation or shell escaping restricts the category to an expected identifier format. Shell metacharacters in a task category can consequently introduce additional commands. The broad `catch` block suppresses errors, which may also make attempted or successful exploitation less visible. This issue becomes especially significant because the documented autonomous flow schedules the learning-direction engine and automatically writes generated tasks into the same queue. ### Attack Path 1. An attacker or another compromised component places a crafted category in `tasks/queue.json`. 2. The user or a scheduled workflow invokes `node engines/learning-direction.js auto` or `discover`. 3. `loadQueue()` parses the persistent queue. 4. `discoverNewSkills()` extracts the malicious task category. 5. The category is concate ...[truncated 692 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Error
Location
engines/learning-direction.js:203
Finding

Execution of an Undeclared and Unpinned Package Through npx

Content
View full analysis
Remediation
View remediation

T02 · Agent Memory Poisoning

Error
Location
engines/evolution-engine.js:99
Finding

Recurring Modification of Shared Agent Memory

Content
View full analysis
Remediation
View remediation

T01 · Skill Instruction Hijacking

Error
Location
engines/skill-creator.js:96
Finding

Unsanitized Workspace Content Embedded Into Generated Skill Instructions

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (53)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 21)May include surrounding context.

md
| **进化引擎** | 任务选择 + 自主执行 | `engines/evolution-engine.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 101)May include surrounding context.

md
| **进化引擎** | 任务选择 + 自主执行 | `engines/evolution-engine.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 104)May include surrounding context.

md
| **进化引擎** | 任务选择 + 自主执行 | `engines/evolution-engine.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 221)May include surrounding context.

md
| **进化引擎** | 任务选择 + 自主执行 | `engines/evolution-engine.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 346)May include surrounding context.

md
| **进化引擎** | 任务选择 + 自主执行 | `engines/evolution-engine.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 352)May include surrounding context.

md
| **进化引擎** | 任务选择 + 自主执行 | `engines/evolution-engine.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 22)May include surrounding context.

md
| **学习提取器** | 从任务提取可复用模式 | `engines/extractor.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 23)May include surrounding context.

md
| **自信度评估** | 计算模式可靠性评分 | `engines/confidence.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 24)May include surrounding context.

md
| **技能创建器** | 高自信模式→可复用技能 | `engines/skill-creator.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 25)May include surrounding context.

md
| **反思引擎** | 每日/每周反思报告 | `engines/reflection.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 107)May include surrounding context.

md
| **反思引擎** | 每日/每周反思报告 | `engines/reflection.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 108)May include surrounding context.

md
| **反思引擎** | 每日/每周反思报告 | `engines/reflection.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 226)May include surrounding context.

md
| **反思引擎** | 每日/每周反思报告 | `engines/reflection.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 231)May include surrounding context.

md
| **反思引擎** | 每日/每周反思报告 | `engines/reflection.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 377)May include surrounding context.

md
| **反思引擎** | 每日/每周反思报告 | `engines/reflection.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 26)May include surrounding context.

md
| **学习方向生成** | 自主发现新方向 + 生成任务 | `engines/learning-direction.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 111)May include surrounding context.

md
| **学习方向生成** | 自主发现新方向 + 生成任务 | `engines/learning-direction.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 236)May include surrounding context.

md
| **学习方向生成** | 自主发现新方向 + 生成任务 | `engines/learning-direction.js` |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 332)May include surrounding context.

md
| **学习方向生成** | 自主发现新方向 + 生成任务 | `engines/learning-direction.js` |

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The document directly instructs users to install the skill and execute init, cron-setup, and startup scripts without any warning, review guidance, or description of side effects. Because these commands can modify the local environment and establish persistent scheduled execution, a user could be induced to run unreviewed code and silently enable ongoing automated behavior on their system.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The document promotes autonomous scheduled execution, auto skill creation, and reflection loops without any explicit warning about what actions may occur automatically, what files may be modified, or how a user can constrain or disable those behaviors. In the context of an agent skill, this can lead users to enable persistent automation they do not fully understand, increasing the risk of unintended file changes, scheduled task registration, or other impactful side effects.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · PUBLISH-SUMMARY.md (reported line 149)May include surrounding context.

bash
# 1. 创建 GitHub 仓库
gh repo create autonomous-learning-cycle --public

# 2. 添加文件
cd skills/autonomous-learning-cycle

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The release notes advertise a 17-minute autonomous loop, automatic task execution, daily/weekly reflection, and automatic skill creation without clearly warning that the system may persistently modify files, consume resources, and change system behavior over time. In a self-modifying or self-extending agent skill, omission of these warnings materially increases the risk that users enable ongoing automation without informed consent or adequate safeguards.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The quick-start instructions tell users to run initialization, cron setup, and start commands, but do not warn that this creates persistent background execution. That can lead users to unknowingly install recurring jobs that continue running, consuming resources and potentially performing autonomous actions long after the initial setup.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The workflow advertises automatic task execution, pattern extraction, and automatic skill creation, but it does not warn that these actions may create or modify local files, configurations, reports, or generated artifacts. Because the skill is explicitly self-modifying in behavior and output generation, failing to disclose filesystem-side effects increases the risk of unreviewed changes and trust in machine-generated artifacts.

Content

No source excerpt is available for this finding.