Back to skill

Security audit

招投标披露全景

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious because its scanners report unpinned installation sources, external service checks that do not match the stated tender-disclosure task, and unclear disclosure before attachment processing may send documents to a third party.

The scanner recommendation is not to install. If you still consider using it, review the external service and attachment-processing steps first, avoid sending sensitive documents or credentials, and verify the installer and source revision before installation. The scan is incomplete because one referenced local path could not be resolved, so it does not establish that the skill is safe.

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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (17)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The supplied chunk performs service discovery and authenticated tool enumeration, not tender-document analysis or production of the eight-field disclosure card. External discovery could support a larger implementation, but this chunk establishes no connection between the discovered tools and the declared task. Its observable activity accesses an otherwise undeclared service and sends a credential to a configurable destination. The code does not establish malicious credential theft, but its demonstrated purpose and resource access do not match the declared workflow.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

The supplied chunk's stated and implemented purpose is to check catalog accessibility and authenticated endpoint acceptance. It does not process a tender notice, extract disclosure fields, attach source anchors, or produce the declared card. Network access and tool discovery could support a larger implementation, but this chunk provides no tender-specific connection establishing that role; its observable purpose is generic MCP connectivity diagnostics. The API key is used as an authentication credential, which alone is not evidence of credential theft.

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 bid-panorama-cn ~/.workbuddy/skills/
# ask the agent for the five-section card defined in SKILL.md, save it as my-bid-card.md
python3 ~/.workbuddy/skills/bid-panorama-cn/scripts/check_bid_card.py ./my-bid-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 bid-panorama-cn ~/.workbuddy/skills/

# 2. 让 agent 按 SKILL.md 的全景卡五段出卡,产物存成我的招标卡.md

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

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The installation command executes an unpinned npm CLI and retrieves skill content without an immutable revision. Although this is a skill installer rather than an MCP server declaration, compromised upstream releases could execute code or introduce malicious agent instructions.

Content

No source excerpt is available for this finding.

Third-party attachment parsing is recommended without disclosing that document contents may leave the user's environment.

Medium
Category
Output Handling
Confidence
87% confidence
Finding

The setup recommends third-party attachment parsing without clearly stating that attachment contents may leave the local environment or requiring approval before upload. Most intended disclosures are public, which reduces risk, but user-supplied documents can contain confidential annotations or other nonpublic information.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The optional dependency installation uses an unversioned executable and a mutable upstream skill collection. A compromised update could introduce executable code or agent instructions, and installing the collection potentially expands the affected surface.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

Selecting one dependency reduces scope but does not pin either the npm installer or the dependency's repository revision. Installation can therefore execute or enable different, potentially compromised content on subsequent runs.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
85% confidence
Finding

The tools/list request is legitimate authentication, not evidence of deliberate exfiltration, but its destination is taken from catalog routing without documented scheme or origin validation. A compromised catalog or incorrectly populated ROUTING_URL could receive the API key, including over plaintext HTTP if permitted.

Content

Scanner excerpt · README.en.md (reported line 52)May include surrounding context.

bash
curl -s "https://cuecue.cn/api/mcp-catalog" | head -c 200
# take routing.url and transport from the response; never hard-code the host
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

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

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The command invokes an unpinned npm installer and obtains the skill from a mutable repository reference. This creates a supply-chain exposure to executable package changes and malicious skill instructions, despite the command's legitimate installation purpose.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The collection installer and its upstream repository are not pinned, so future invocations can execute or install unreviewed changes. The interactive selection described here limits which skills are installed but does not protect against compromised installer or repository updates.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The single-skill selector does not make the npm executable or repository content immutable. A compromised release could run code during installation or introduce malicious instructions into the optional data or parsing dependency.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
85% confidence
Finding

Sending a bearer token for tools/list is expected service authentication, but the example attaches it to a catalog-derived URL without specifying destination validation. If that routing value is compromised or misconfigured, the credential can be disclosed to an unintended host or transmitted over an insecure scheme.

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 recovery step lacks a warning about sending document contents to a third-party service.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The recovery action recommends attachment parsing without an explicit external-transfer warning or authorization gate. Public tender disclosures normally have limited confidentiality, but locally supplied attachments may include private annotations or other information unsuitable for third-party processing.

Content

No source excerpt is available for this finding.

The attachment-parsing workflow lacks an explicit warning that document contents are sent to an external service.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The workflow directs long attachments through a server-backed parsing channel without explicitly disclosing document transmission or requiring approval before external processing. Its focus on public notices reduces typical sensitivity, but this instruction could still transfer nonpublic user-supplied attachment contents without informed authorization.

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 warning that document contents are sent to an external service.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The attachment workflow delegates parsing to an external service without explicitly requiring disclosure or authorization before transmitting document contents. Public tender notices reduce the usual sensitivity, but user-supplied attachments or summaries may include nonpublic bid information or personal data, creating a conditional confidentiality risk.

Content

No source excerpt is available for this finding.