Back to skill

Security audit

Mixlab Daily Briefing( 高价值内容筛选)

Security checks for vulnerabilities and agentic risk

Overview

This skill does what it says, but it lets AI-generated output drive authenticated updates to Mixdao records without enough final validation.

Review before installing if the Mixdao token can update important or public content. Use a narrowly scoped Mixdao token, check generated markdown and recommendations before submission where possible, and add a final allowlist validation so only fetched cachedStoryIds can be patched.

Vulnerability Patterns
  • 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
  • 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
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/02-briefing.js:315
Finding

Unvalidated Model-Generated IDs Can Cause Unauthorized Recommendation Updates

Content
View full analysis
({ id: it.id, recommendationText: (it.recommendationText || it.recommendation || '').trim(), })).filter((it) => it.id && it.recommendationText); for (const it of items) flat.push({ id: it.id, title: filledGroups[i]?.items?.find((x) => x.id === it.id)?.translatedTitle || filledGroups[i]?.items?.find((x) => x.id === it.id)?.title, recommendationText: it.recommendationText, }); ``` ```js const batchBody = toSubmit.map(({ id, recommendationText }) => ({ cachedStoryId: id, recommendationText, })); ``` ### Technical Analysis Story titles and text obtained from the external Mixdao feed are embedded in the MiniMax prompt. This content must therefore be treated as untrusted input and may contain prompt-injection instructions. The grouping stage validates model-returned IDs against a set of expected IDs. However, the subsequent summary and recommendation stage accepts every nonempty `id` returned by the model without verifying that: - The ID exists in the original fetched item set. - The ID belongs to the group submitted to the model. - The ID appears only once. - The returned IDs exactly correspond to the expected IDs. The accepted values are copied into `toSubmit`, converted directly into `cachedStoryId` fields, and sent through an authenticated Mixdao PATCH request. Consequently, the mod ...[truncated 1918 chars]
Remediation
View remediation
group.items.map((item) => String(item.id))) ); const seenIds = new Set(); const batchBody = toSubmit.map(({ id, recommendationText }) => { const normalizedId = String(id); if (!expectedIds.has(normalizedId)) { throw new Error(`Model returned an unauthorized item ID: ${normalizedId}`); } if (seenIds.has(normalizedId)) { throw new Error(`Model returned a duplicate item ID: ${normalizedId}`); } if (typeof recommendationText !== 'string' || !recommendationText.trim()) { throw new Error(`Model returned an invalid recommendation for: ${normalizedId}`); } seenIds.add(normalizedId); return { cachedStoryId: normalizedId, recommendationText: recommendationText.trim(), }; }); ``` 6. Apply equivalent per-group validation so an ID cannot be reassigned between groups. 7. Delimit external story content clearly in prompts and explicitly state that instructions contained in story data must not be followed. This is defense in depth and must not replace deterministic output validation. 8. Where supported, use schema-constrained model output and enforce the schema locally. 9. Configure the Mixdao API token with the narrowest server-side update permissions available so that failed client-side validation cannot ...[truncated 41 chars]
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (16)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

声明描述的是一个完整的“抓取→AI分类分组→生成摘要/推荐语→回写提交到 mixdao”的流水线技能;而给出的代码片段只覆盖最前面的数据获取与本地缓存步骤。它访问的资源与声明中的 mixdao latest 接口一致,这部分是匹配的,但主功能存在重大缺失:没有模型调用,没有内容分组或摘要,没有推荐语生成,也没有任何向 mixdao 回传数据的网络请求。因此代码实际行为与声明的主要目的 materially different,应判定为不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

主要不匹配点在于数据来源:描述写的是从 mixdao latest 获取数据,但该脚本仅作为“步骤2”,从命令行参数指定的本地 JSON 文件读取数据,没有任何获取 latest 列表的逻辑;这是对资源访问和能力边界的实质性差异。其余大部分行为与描述基本一致,包括使用 Anthropic SDK 指向 MiniMax-M2.5、生成分组摘要和推荐语、按 cachedStoryId 批量 PATCH 推荐语、以及上传简报到 mixdao。另有次要差异:代码将分组重试 loop 限制为3次;并且当 items 少于12条时,把每组最少条数从3放宽到2,这与“每组至少3条”的固定描述不完全一致。综合看,应判定为描述与代码存在实质不匹配。

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 13)May include surrounding context.

md
| `scripts/02-briefing.js` | 读取步骤 1 的 temp 文件 → 调用 **MiniMax-M2.5** 做分组(agent loop 直至满足约束)→ 生成分组摘要(100 字内)与每条推荐语(140 字内,创业者视角)→ 保存简报 MD,并 **PATCH /api/briefing*

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 32)May include surrounding context.

md
| `scripts/02-briefing.js` | 读取步骤 1 的 temp 文件 → 调用 **MiniMax-M2.5** 做分组(agent loop 直至满足约束)→ 生成分组摘要(100 字内)与每条推荐语(140 字内,创业者视角)→ 保存简报 MD,并 **PATCH /api/briefing*

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 38)May include surrounding context.

md
| `scripts/02-briefing.js` | 读取步骤 1 的 temp 文件 → 调用 **MiniMax-M2.5** 做分组(agent loop 直至满足约束)→ 生成分组摘要(100 字内)与每条推荐语(140 字内,创业者视角)→ 保存简报 MD,并 **PATCH /api/briefing*

Known Vulnerable Dependency: form-data==4.0.5 — 1 advisory(ies): CVE-2026-12143 (form-data: CRLF injection in form-data via unescaped multipart field names and f)

High
Category
Supply Chain
Confidence
91% confidence
Finding

The lockfile includes form-data 4.0.5, which is flagged with a HIGH advisory for CRLF injection in multipart field names. Because this skill fetches external content and uploads a briefing to mixdao, any code path that builds multipart form data from untrusted story metadata or model-generated text could let an attacker inject crafted part headers or corrupt the outbound request. The issue is transitive rather than obviously exploited here, so the package presence is a real supply-chain vulnerability, but exploitability depends on whether multipart field names or filenames are influenced by untrusted input.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill declares access to sensitive environment variables (MIXDAO_API_KEY, ANTHROPIC_API_KEY) but does not define any explicit tool scope or permission boundary. That increases the blast radius because an executing agent may expose or misuse secrets without a clear least-privilege contract, especially since the workflow also performs outbound API calls and file writes.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 53)May include surrounding context.

md
const apiKey = process.env.ANTHROPIC_API_KEY;
    if (!apiKey) throw new Error('ANTHROPIC_API_KEY is not set.');
    return new Anthropic({
        baseURL: baseURL || 'https://api.minimaxi.com/anthropic',
        apiKey,
    });
}

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · scripts/02-briefing.js (reported line 110)May include surrounding context.

js
const apiKey = process.env.ANTHROPIC_API_KEY;
    if (!apiKey) throw new Error('ANTHROPIC_API_KEY is not set.');
    return new Anthropic({
        baseURL: baseURL || 'https://api.minimaxi.com/anthropic',
        apiKey,
    });
}

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The script uploads generated markdown content to /api/briefing and submits recommendation text batches to /api/latest/recommendation, but there is no confirmation prompt or normal user-facing notice before these network transmissions occur. Although comments describe the behavior, the runtime path lacks a clear disclosure to the operator that local file content and generated text will be sent to external services.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
86% confidence
Finding

The dependency uses a caret range, which allows npm to install newer compatible releases automatically. This can introduce supply-chain risk if a future upstream release is compromised or contains a breaking security regression, especially because this skill processes external data and uses API credentials.

Content

Scanner excerpt · package.json (reported line 7)May include surrounding context.

json
"type": "module",
    "private": true,
    "dependencies": {
        "@anthropic-ai/sdk": "^0.32.1",
        "jsonrepair": "^3.13.2"
    }
}

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
86% confidence
Finding

The dependency is not fully pinned and may resolve to different versions over time, weakening build reproducibility and increasing exposure to malicious or vulnerable upstream releases. In this skill, any compromised library could affect processing of untrusted content or handling of secrets used to call external services.

Content

Scanner excerpt · package.json (reported line 8)May include surrounding context.

json
"private": true,
    "dependencies": {
        "@anthropic-ai/sdk": "^0.32.1",
        "jsonrepair": "^3.13.2"
    }
}

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The comment and DATE_TZ constant force date handling to Asia/Shanghai, which is a locale-specific behavior. There is no visible opt-in, configuration option, or justification that this skill is intended only for a China-specific context.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The script reads MIXDAO_API_KEY from the environment and uses it in an Authorization header for an outbound HTTPS request. Although the header comment documents the requirement, there is no runtime confirmation or user-facing warning before transmitting credentials to an external service.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The code accesses MIXDAO_API_KEY and later ANTHROPIC_API_KEY from the environment to authenticate outbound requests, but provides no user-facing warning that credentials will be consumed for external service calls. Error handling exists when variables are missing, yet there is no disclosure of this sensitive credential usage during normal execution.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The script creates or overwrites a .md file derived from the input path using fs.writeFileSync, but does not notify the user beforehand that a local file will be written. A success message is printed only after completion, which does not serve as advance disclosure for the file write operation.

Content

No source excerpt is available for this finding.