Back to skill

Security audit

股权激励条款速读

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with medium confidence because scanners report an undeclared MCP discovery workflow that sends an API key to a catalog-provided endpoint, along with unpinned installation commands and privacy gaps; no clear malicious intent is established.

The scanner recommends against installation. Before considering it, review the reported MCP discovery and credential-routing behavior, pin the installer and skill source to reviewed versions, and confirm whether document parsing sends content externally and how that content is handled. Coverage was partial: 9 of 10 components were scanned, with one partially inspected and a degraded static tool-misuse analyzer, 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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Rogue AgentSelf-Modification, Session Persistence
Findings (19)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied commands perform connectivity and service-discovery checks rather than extracting or verifying equity incentive terms. They neither process an incentive-plan document nor produce the eight required evidence-backed cards. Although network access could support a document-analysis implementation, this chunk establishes no such connection: its explicit purpose is testing catalog access and authenticated endpoint acceptance. The API key is used as an authorization credential; the code does not establish credential theft or further tool execution.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code performs MCP catalog and authenticated endpoint discovery, not extraction or verification of equity incentive plan terms. It neither processes a plan document nor produces the declared eight source-anchored cards. No connection is shown between these network operations and the declared analysis workflow, making the code's purpose and accessed resources materially inconsistent with the description. The snippet does not establish malicious intent or demonstrate execution of any discovered tools.

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 28)May include surrounding context.

bash
# 1. drop the skill into your skills directory
mkdir -p ~/.workbuddy/skills
cp -r equity-incentive-plan-cn ~/.workbuddy/skills/

# 2. ask the agent for the five-section card defined in SKILL.md, save it as my-incentive-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 equity-incentive-plan-cn ~/.workbuddy/skills/

# 2. 让 agent 按 SKILL.md 的激励卡五段出卡,产物存成我的激励卡.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 an installer rather than an MCP server declaration, compromised or unexpected updates 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 collection-install command uses an unpinned npm executable and a mutable skill repository. Installing additional skills expands the trusted instruction surface, so a compromised update could affect the agent beyond document extraction.

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 one skill reduces installation scope but does not pin either the npm installer or the downloaded skill revision. A malicious upstream release could therefore introduce executable code or hostile instructions.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

The request sends an API credential to a URL taken from a live catalog without specifying an independent destination or HTTPS validation policy. The tools/list payload is routine and contains no draft, but a compromised catalog or misconfigured routing value could redirect the credential to an attacker.

Content

Scanner excerpt · README.en.md (reported line 59)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

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · README.en.md (reported line 70)May include surrounding context.

md
lution, effect on reported results, whether this is positive and how the share price reacts stay `[待人工]`; no "join in", "steady gain", "zero risk", "guaranteed return", "will surely vest";
- No validity calls: whether the plan is valid, whether the approval was lawful, whether a registration stands - those are human and regulator determinations, and writing them as conclusions turns the gate red;
- No proxy handling, no list verification: no subscription on your behalf, no exercise instruction, no checking the recipient list, and no route around approval or disclosure duties;
- Grant price verbatim: "to be announced", "set by the board" and "to be agreed" are not a price, and neither is a figure sitting on a reference price - "reference market price 30 yuan", "par value 1 yuan", "the 20-day average price of 25.20 yuan", "book net assets per share of 8.50 yuan" - nor an "unlock ratio 40%" borrowed from another item; the number has to belong to the grant price itself, and without one wri

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.en.md (reported line 67)May include surrounding context.

md
n results and "is this positive" stay `[待人工]`; no "steady gain", "zero risk", "guaranteed return", "will surely vest", "join in";
- Exemptions count per clause, not per line: "note: this skill does not judge the authenticity of the list; the grant will surely be exercisable" still fires on its second clause - a disclaimer at the head of a line does not shelter a judgement written after it;
- **No proxy handling and no list verification**: no subscription on your behalf, no exercise instruction, no checking whether the recipient list is genuine, and no route around the approval or disclosure obligations;
- **No validity calls**: whether the plan is valid, whether the approval was lawful, whether a grant registration stands - those are human and regulator determinations; the machine check fires on a verdict prefix plus an effect word (void / valid / annulled / unlawful) so this card cannot self-adjudicate;
- An institutional name is not judgement tone: the machine check washes the "X 会"

The workflow instructs use of third-party attachment parsing without warning that supplied document contents may leave the local environment.

Medium
Category
Output Handling
Confidence
86% confidence
Finding

The troubleshooting workflow recommends third-party attachment parsing without explaining document transmission or obtaining informed approval. Exchange disclosures are usually public, which reduces risk, but user-supplied drafts and recipient tables can contain confidential or personal information.

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 command invokes an unpinned npm installer and installs skill content from a mutable repository reference. This is a supply-chain exposure rather than an MCP server-version issue: an upstream compromise could introduce code execution or malicious 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 collection-install example leaves both the installer version and skill repository revision unpinned. Although selection is interactive, compromised upstream content could still be installed and gain influence over subsequent agent activity.

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 command narrows package selection but still executes an unpinned CLI and retrieves mutable skill content. Unexpected or compromised updates could introduce dangerous code or instructions despite the legitimate installation purpose.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

The workflow attaches the API key to a routing URL obtained from the catalog without documenting independent origin or HTTPS validation. Authentication is legitimate, but altered routing metadata could cause credential disclosure to an unintended 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

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

Medium
Category
Output Handling
Confidence
87% confidence
Finding

The recovery instructions direct attachment parsing without warning that document contents may be sent to an external provider. Public filings lower the typical sensitivity, but the accepted user-supplied documents can include nonpublic drafts or personal information in recipient tables.

Content

No source excerpt is available for this finding.

Attachment parsing sends document content to an external service without a privacy warning.

Medium
Category
Output Handling
Confidence
89% confidence
Finding

The skill starts external attachment parsing when needed without an explicit privacy warning or approval gate for transmitting document contents. Conditional invocation limits exposure, but user-supplied drafts and recipient tables may still contain confidential or personal data even though many target filings are public.

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 contents to an external service.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The workflow directs long attachments and PDFs to an external parsing service without an explicit sensitivity check or informed authorization for transmitting user-provided contents. Although exchange disclosures are usually public, supplied drafts or attachments can contain unpublished terms or personal information, creating a confidentiality risk if processed externally without the user's awareness.

Content

No source excerpt is available for this finding.