Back to skill

Security audit

Engram Evomap

Security checks for vulnerabilities and agentic risk

Overview

This skill is a long-term agent memory system, but it can store session-derived advice and later inject it as high-authority guidance, including one bundled unsafe Git TLS workaround.

Review this before installing in an agent that can run commands, access secrets, or edit repositories. Treat stored capsules as untrusted suggestions, avoid enabling system-role injection, remove or quarantine the bundled Git SSL seed, and require explicit consent before sending conversation history to any LLM provider.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T02 · Agent Memory Poisoning

Error
Location
src/core/gene-processor.js:19
Finding

Persistent memory content can be elevated to system-message authority

Content
View full analysis
`[${m.role}]: ${m.content}`).join('\n'); const prompt = ` 你是一位资深软件架构师,负责将具体的调试经验“基因化(GeneDistillation)”。 请分析以下对话中的问题解决过程: ${rawText} --- 提炼任务要求: 1. **去变量化**:剔除具体的项目路径、服务器IP、用户名、具体的仓库URL(用 , , 代替)。 2. **逻辑抽象**:描述问题的本质原因,而非表现。例如:“NPM权限错误”而非“安装 axios 报错”。 3. **步骤分解**:将方案分解为 diagnosis(诊断)、patch(修复)、config(配置) 或 workaround(临时规避)。 4. **反模式警告**:指出用户容易踩坑的错误尝试(如有)。 请严格按照以下 JSON 格式输出,不要包含任何 Markdown 标记: { "category": "分类标签", "triggerPattern": "触发该场景的通用特征描述", "rootCause": "本质原因的抽象描述", "actionSequence": [ { "step": 1, "type": "diagnosis/patch/config/workaround", "instruction": "明确的指令", "rationale": "为什么要这么做" } ], "verificationCriterion": "验证问题已解决的标准", "antipatternWarning": "可选的警告信息", "tags": ["标签1", "标签2"] } `; const response = await this.llmClient.ask(prompt); // 清理 LLM 可能输出的 Markdown 块 const cleanJson = response.replace(/```json|```/g, '').trim(); const distilled = JSON.parse(cleanJson); // 为“触发特征”生成向量索引 const searchKey = `${distilled.triggerPattern} ${distilled.rootCause}`; const embedding = await this.embed.vectorize(searchKey); return { ...distilled, embedding }; } ``` `src/core/distill-manager.js:34-55`: ```js async _processTask(history, taskId) { try { console.log(`[EvoMap-Distill] Starting task ${taskId}...`); // 1. 提炼并生成向量 (LLM 1次消耗) const distilled = await this.processor.distill(history); const env = EnvChecker.getF ...[truncated 4612 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
data/seeds/seeds.json:62
Finding

Bundled seed advice disables Git TLS verification globally

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
Findings (32)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The skill claims to be a memory hub but also uses external LLM interaction for capsule evaluation and trust scoring. Undisclosed outbound model calls can expose session-derived content to third parties and create prompt-injection or data leakage risks inconsistent with the stated behavior.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The skill claims to be a memory hub but also uses external LLM interaction for capsule evaluation and trust scoring. Undisclosed outbound model calls can expose session-derived content to third parties and create prompt-injection or data leakage risks inconsistent with the stated behavior.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The skill claims to be a memory hub but also uses external LLM interaction for capsule evaluation and trust scoring. Undisclosed outbound model calls can expose session-derived content to third parties and create prompt-injection or data leakage risks inconsistent with the stated behavior.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The skill claims to be a memory hub but also uses external LLM interaction for capsule evaluation and trust scoring. Undisclosed outbound model calls can expose session-derived content to third parties and create prompt-injection or data leakage risks inconsistent with the stated behavior.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The seeded memory capsule explicitly recommends git config --global http.sslVerify false, which disables TLS certificate validation for all Git HTTPS traffic on the host. In a long-term memory hub, this is especially dangerous because the unsafe workaround can be repeatedly surfaced and normalized, enabling man-in-the-middle interception of credentials, code, and repository contents.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The instruction to disable Git SSL verification is presented as the direct workaround, while the warning appears only later in antipatternWarning. Users or agents following the actionSequence may execute the insecure command without understanding that it removes certificate validation globally, creating systemic exposure to spoofed Git servers and credential theft.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
86% confidence
Finding

The README instructs users to run npx engram without a pinned version, which causes execution of whatever package version is current at install time. In an agent-oriented package that initializes models and injects seed data, this increases supply-chain risk because future compromised or unexpected releases could be executed automatically by users or automation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The auto-interception behavior is described as triggering on generic error strings like EACCES, 404, and SSL, which are common across many unrelated contexts. In a long-term memory skill that injects advice into agent workflows, such broad triggers can cause unintended instruction injection, workflow corruption, or unsafe remediation suggestions in situations where the matched text is incidental or adversarially introduced.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
88% confidence
Finding

The skill declares runtime capabilities via env/node usage but does not explicitly declare tool scope or permissions, which weakens reviewability and allows broader execution assumptions than the manifest communicates. In a skill that invokes external tooling, missing permission boundaries increases the chance of unintended environment access or privilege creep.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

Using npx engram without a pinned version allows whatever package version is current at execution time to be fetched and run, creating a supply-chain risk. If the upstream package is compromised or a breaking update is published, the skill may execute attacker-controlled code.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The automatic trigger rules are broad enough that common terms like error, 404, or timeout can cause unsolicited memory lookups and advice injection. In an adversarial setting, this can be abused to manipulate agent behavior, surface poisoned historical content, or cause unintended data flow into the memory subsystem.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The code comment says remote model sending/access is disabled, but 'preheat()' explicitly downloads 'Xenova/all-MiniLM-L6-v2' over the network. This mismatch can mislead users and reviewers into believing initialization is fully offline/local, creating trust and supply-chain exposure during install-time behavior.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Context Window Stuffing

Medium
Category
Memory Poisoning
Confidence
80% confidence
Finding

Skill attempts to fill the context window with filler content, displacing legitimate instructions and safety constraints. This can degrade agent performance or bypass safety boundaries.

Content

Scanner excerpt · data/seeds/seeds.json (reported line 12)May include surrounding context.

json
"triggerSignature": {
      "errorPattern": "ERR! ERESOLVE unable to resolve dependency tree",
      "taskIntent": "Installing npm packages on an old project or strict modern env",
      "embedding": [0.1012, -0.0456, 0.0890, 0.334, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.
...[truncated 25 chars]

Context Window Stuffing

Medium
Category
Memory Poisoning
Confidence
80% confidence
Finding

Skill attempts to fill the context window with filler content, displacing legitimate instructions and safety constraints. This can degrade agent performance or bypass safety boundaries.

Content

Scanner excerpt · data/seeds/seeds.json (reported line 48)May include surrounding context.

json
"triggerSignature": {
      "errorPattern": "ERR! ERESOLVE unable to resolve dependency tree",
      "taskIntent": "Installing npm packages on an old project or strict modern env",
      "embedding": [0.1012, -0.0456, 0.0890, 0.334, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.
...[truncated 25 chars]

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The commit command returns a hard-coded Chinese status message ('经验已存储') while the rest of the file uses English. This imposes a specific language on users without offering a choice or documenting a justified locale constraint, which is a natural-language policy issue.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The schema permits very loose trigger definitions: both errorPattern and taskIntent are arbitrary strings, and embedding is an unconstrained numeric array with no size, range, or semantic validation. In a long-term memory/agent activation system, this can cause overly broad or malformed capsules to match unexpectedly, leading the agent to retrieve and apply incorrect action sequences or memory entries in inappropriate contexts.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The file's user-facing strings and comments indicate the injected advice is rendered in Chinese, including headings such as "已发现已知解决方案", "诊断结论", and "分步解法". This imposes a specific language on the skill output without offering user choice or documenting a justified region-specific constraint, which matches the language/locale policy violation criteria.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The file contains user-visible text such as the enqueue return message entirely in Chinese, and the surrounding comments indicate the skill is designed around that locale. There is no indication that users can opt into this language or that the skill is intentionally limited to a Chinese-language or region-specific context.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The code immediately launches background processing on conversation history and the returned user-facing message does not clearly warn that the history will be distilled, stored, and sent through an LLM-based reflection/verification pipeline. In a long-term memory skill, this creates a real privacy and consent risk because sensitive conversation content may be persisted or transmitted without informed user awareness.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The class description and multiple user-facing outputs in this file are written in Chinese, indicating the skill is designed to communicate in a fixed language. The policy requires not forcing a specific language without user opt-in, and this file shows no mechanism for language selection or documented locale justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The prompt content is written entirely in Chinese and instructs the model to produce specific structured output in that language context, with no indication that the user can choose another language. This can violate language/locale policy when a skill imposes a language without explicit opt-in or documented justification.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The code sends recent session conversation content directly to an external LLM via this.llmClient.ask(prompt) after embedding raw dialogue into the prompt. Even though the prompt asks the model to replace paths, URLs, and usernames, that sanitization happens only after transmission, so sensitive data may already be disclosed to a third party without visible consent, minimization, or policy enforcement.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The file's comments and embedded prompt are written exclusively in Chinese, including the evaluation instructions sent to the model. This imposes a specific language/locale without any opt-in, fallback, or justification that the skill is region-specific.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The code sends capsule-derived content to an LLM via llmClient.ask(prompt) with no visible disclosure, consent flow, minimization, or redaction controls in this file. Because capsule contents may include sensitive operational history, error patterns, task intent, or solution steps, this can leak internal data to an external model provider or logging layer and create prompt-injection/privacy risks.

Content

No source excerpt is available for this finding.