Back to skill

Security audit

债券条款卡

Security checks for vulnerabilities and agentic risk

Overview

The skill is rated suspicious with medium confidence: scanners report undeclared external service and credential use, unpinned installation sources, and document privacy and citation-check concerns; the scan was partial.

Treat this as a review-level warning, not a finding that the author is malicious. Before installing, consider whether the external MCP checks are needed, how credentials and document contents would be handled, and whether the installer and source can be pinned and reviewed. The AIG scan completed without findings, but Skillspector reported partial analysis because a referenced path was missing, so the evidence is incomplete.

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
  • 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
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (17)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied shell commands test catalog access and authenticated MCP tool listing rather than extract or cite bond terms. They introduce external service discovery and credential use that the declared purpose does not describe. No connection is shown between these operations and processing a bond document, so they cannot be established as supporting implementation details. The snippet does not establish credential theft or any subsequent tool execution.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

代码的实际用途是 MCP 服务发现与认证连通性检查,没有读取募集说明书、提取八项债券条款、生成原句锚点或标注缺项的逻辑。虽然网络访问可以作为某些技能的辅助实现,但该片段未体现其与债券条款提取的关联,实际主要行为及所访问资源与声明用途存在实质差异。

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 · README.en.md (reported line 27)May include surrounding context.

3. Install and run

bash
mkdir -p ~/.workbuddy/skills
cp -r bond-terms-cn ~/.workbuddy/skills/
# ask the agent for the five-section card defined in SKILL.md, save it as my-bond-card.md
python3 ~/.workbuddy/skills/bond-terms-cn/scripts/check_bond_card.py ./my-bond-card.md

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 · README.md (reported line 28)May include surrounding context.

bash
# 1. 放到技能目录
mkdir -p ~/.workbuddy/skills
cp -r bond-terms-cn ~/.workbuddy/skills/

# 2. 让 agent 按 SKILL.md 的债券卡五段出卡,产物存成我的债券卡.md

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The installation command runs an unpinned npm CLI and retrieves skill content from a mutable repository reference. Although this is not itself an MCP server declaration, compromised or unexpectedly changed upstream releases could execute code or install adversarial agent instructions.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The optional integration installation uses an unpinned executable and mutable upstream skill collection. This creates a supply-chain exposure, potentially affecting integrations that subsequently handle credentials and documents; it is not evidence that the current upstream content is malicious.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

Selecting a single skill narrows the installation but does not pin either the npm executable or the downloaded skill revision. Upstream compromise or changes could therefore introduce code or instructions outside the reviewed version.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

The request intentionally authenticates to an external service, but its destination comes from catalog-provided routing without a documented HTTPS or trusted-origin check. A compromised catalog or incorrectly populated routing variable could cause the bearer key to be sent to an unauthorized endpoint; the tools/list body itself contains no bond document.

Content

Scanner excerpt · README.en.md (reported line 55)May include surrounding context.

2) take routing.url and transport for the domain from the response; never hard-code the host

3) with the key, send tools/list - proves that endpoint accepted the key, nothing else

curl -s -X POST "$ROUTING_URL" -H "Authorization: Bearer $CUE_API_KEY"
-H "Content-Type: application/json" -H "Accept: application/json, text/event-stream"
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | head -c 400

text

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The command executes an unpinned npm CLI and installs skill content from a mutable repository reference. This is a supply-chain weakness rather than proof of a malicious MCP server: upstream changes could introduce executable code or adversarial instructions.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The collection installation leaves both the npm CLI and upstream skill revision unpinned. Interactive selection does not prevent a compromised installer or changed integration from introducing unsafe behavior, particularly where integrations handle API credentials and document contents.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The single-skill flag limits selection but still downloads mutable integration content using an unpinned npm executable. An upstream compromise or unreviewed update could change the agent's instructions or integration behavior.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

The diagnostic sends a bearer credential to a catalog-derived routing URL without documenting destination validation. Authentication is legitimate here, but a compromised routing response or incorrectly set variable could direct the key to an unauthorized host; no document content is present in this request.

Content

Scanner excerpt · README.md (reported line 59)May include surrounding context.

2) 域路由现值:从返回里取该域的 routing.url 与 transport,别把域名抄死

3) 带 key 实连后发 tools/list——只证该数据腿端点被接受,不证别的入口

curl -s -X POST "$ROUTING_URL" -H "Authorization: Bearer $CUE_API_KEY"
-H "Content-Type: application/json" -H "Accept: application/json, text/event-stream"
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | head -c 400

text

The attachment-parsing workflow lacks a warning about sharing document contents with Cue.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The workflow recommends external attachment parsing without explicitly warning that document contents may leave the user's environment or requiring upload approval. Public prospectuses reduce the usual sensitivity, but user-supplied issuance drafts or marketing materials may be confidential; the statement that this skill does not persist files does not establish the external provider's retention behavior.

Content

No source excerpt is available for this finding.

The attachment-parsing workflow lacks a warning that user-supplied document content is sent to an external service.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The skill directs attachments through an external parsing channel when needed but does not require informed approval before transferring user-supplied content. Mentioning server billing and availability does not explain the privacy implications, and a supplied bond document may be a confidential draft rather than a public prospectus.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: file_read, file_write.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The attachment-parsing workflow lacks a privacy warning about sending document content to an external service.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The workflow initiates external attachment parsing when quotes reside in attachments, without an explicit privacy warning or pre-transfer authorization requirement. Limiting parsing to necessary sections reduces scope but does not prevent disclosure of confidential issuance drafts or other nonpublic user-supplied material.

Content

No source excerpt is available for this finding.

The rating check accepts an empty or unrelated anchor label instead of enforcing the manifest’s same-sentence citation requirement.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The rating check only tests whether the same line contains an anchor label; it does not require a nonempty locator or ensure that the anchor belongs to the sentence containing the rating. Consequently, a non-item line such as “主体评级 AAA。锚:” passes this check, allowing unsupported ratings to appear validated. This matters in a bond-analysis skill because traceable ratings are an explicit safeguard against misleading financial information, although this flaw does not enable code execution or data exfiltration.

Content

No source excerpt is available for this finding.