Back to skill

Security audit

公告速览·A股

Security checks for vulnerabilities and agentic risk

Overview

The scanners report risks around sending credentials to a dynamically chosen service and using unpinned installation sources, so the verdict is suspicious; this indicates review-worthy weaknesses, not proof of malicious intent.

Consider avoiding installation until the API key destination is independently verified and the installer and source are pinned to reviewed versions. Check what company or watchlist data would be sent externally and what file access the host grants. The scan was partial: all eight listed components were scanned, but a local path-like reference in SKILL.md could not be resolved, 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 (13)

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 29)May include surrounding context.

bash
# 1. drop it into the skills directory
mkdir -p ~/.workbuddy/skills
cp -r announcement-daily-cn ~/.workbuddy/skills/

# 2. ask the agent for the four-section list defined in SKILL.md §3, save it as my-digest.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 announcement-daily-cn ~/.workbuddy/skills/

# 2. 让 agent 按 SKILL.md 的 §3 四段出清单,产物存成我的速览.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 installs skill content from a mutable repository reference. A compromised dependency or upstream update could execute code or introduce malicious agent instructions; this is a supply-chain risk rather than evidence of an intentionally malicious MCP server.

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 whole-set installation invokes an unpinned npm CLI and obtains selectable skills from a mutable repository. Compromise of either upstream could introduce executable code or persistent agent instructions, and the broad selection workflow increases the amount of content requiring review.

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 reduces scope but does not pin the npm installer or the source revision. An upstream compromise could therefore change the installed retrieval or parsing instructions without any change to this documentation.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

Bearer authentication is expected for this service, but the destination is taken from a live catalog without an explicit HTTPS or authorized-origin validation step. If that routing value is compromised or incorrectly supplied, the command sends CUE_API_KEY to an unintended recipient; the catalog itself is initially fetched over HTTPS, which reduces ordinary network-interception risk.

Content

Scanner excerpt · README.en.md (reported line 60)May include surrounding context.

2) take the domain's routing.url and transport from that response; do not copy the host into a config by hand

3) with a key, send tools/list — proves only that this data-leg endpoint accepts the key, and proves nothing about other entries

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 uses an unpinned npm installer and a mutable repository reference for the skill. A compromised release or repository update could run code or install malicious agent instructions, although this is not specifically an MCP-server version reference.

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 whole-set installer retrieves mutable skill content using an unpinned CLI. Upstream compromise could introduce malicious code or instructions, and selecting additional skills expands the trusted dependency set.

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 option limits installation scope but leaves both the CLI version and repository revision mutable. An upstream compromise could alter the retrieval or parsing skill subsequently trusted by the agent.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

The connectivity test sends a bearer credential to a URL obtained dynamically from the catalog without specifying destination validation. Authentication is legitimate, but a compromised or mistaken routing value could disclose the key to another origin or an insecure endpoint.

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

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.

External retrieval workflow lacks a privacy warning about transmitting user-supplied company or watchlist queries.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The workflow routes user-supplied company or watchlist queries through an external service without explicitly explaining that disclosure or offering a local alternative at the point of transmission. Although individual filing queries concern public companies, an aggregated watchlist can reveal confidential investment interests or research activity.

Content

No source excerpt is available for this finding.

The self-test enumerates the shared temporary directory despite explicitly claiming not to read directories.

Low
Category
Excessive Agency
Confidence
98% confidence
Finding

The self-test enumerates the shared temporary directory and treats any matching check-digest-* entry as a cleanup failure, including entries belonging to other runs or users. This exceeds its stated directory-access boundary and permits unrelated or deliberately created entries to cause false failures; it does not read their contents or delete those unrelated entries.

Content

No source excerpt is available for this finding.