Back to skill

Security audit

GitMind思乎平替迁移:AI-HIVE多模型工作流

Security checks for vulnerabilities and agentic risk

Overview

This skill is not clearly malicious, but it needs Review because a narrow migration helper includes broad authenticated remote AI-HIVE tool-calling behavior.

Review before installing. Only use this skill if you intend to connect to AI-HIVE, understand that credentials may be used for remote MCP calls, and are comfortable with a helper that can invoke any AI-HIVE tool name you supply. Avoid setting AI_HIVE_MCP_URL unless you fully trust the endpoint, prefer OAuth or a scoped/revocable API key, and confirm budget and data sensitivity before any paid generation or upload.

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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (14)

Tainted flow: 'request' from os.environ.get (line 67, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Confidence
90% confidence
Finding

Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.

Content

Scanner excerpt · scripts/ai_hive_mcp.py (reported line 28)May include surrounding context.

python
def fetch_json(url: str) -> dict:
    request = urllib.request.Request(url, headers={"accept": "application/json"})
    with urllib.request.urlopen(request, timeout=20) as response:
        return json.loads(response.read().decode("utf-8"))

Tainted flow: 'request' from os.environ.get (line 67, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Confidence
92% confidence
Finding

The script allows the MCP endpoint to be overridden via AI_HIVE_MCP_URL and then automatically sends authentication material from environment variables to that endpoint in post(). If a user runs this tool in an environment where AI_HIVE_MCP_URL is attacker-controlled or misconfigured, the script can exfiltrate the API key or bearer token to an arbitrary remote server, making this an SSRF-plus-secret-disclosure issue.

Content

Scanner excerpt · scripts/ai_hive_mcp.py (reported line 74)May include surrounding context.

python
method="POST",
    )
    try:
        with urllib.request.urlopen(request, timeout=60) as response:
            result = parse_payload(response.read(), response.headers.get("content-type", ""))
            return result, response.headers.get("mcp-session-id") or session_id
    except urllib.error.HTTPError as error:

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared purpose is a narrow GitMind migration-evaluation workflow, but the content instructs use of a generic MCP client that can list all tools, perform arbitrary tool calls, inspect OAuth metadata, and invoke potentially paid remote actions. That capability expansion is dangerous because users may trust the narrow business framing while the actual operational scope enables broader remote access and side effects than disclosed.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/mcp-binding.md (reported line 63)May include surrounding context.

md
## Codex、Claude 与 ChatGPT

优先使用各客户端的远程 MCP/连接器界面;若支持 JSON 配置,使用 OAuth 最小配置。客户端必须支持 Streamable HTTP、OAuth 2.1、PKCE 和动态客户端注册。若出现 `unauthorized_client`,保留脱敏错误和客户端版本,通过 AI-HIVE 登录后的客服渠道反馈,不要把 access token 粘贴到聊天中。

## Gemini 与支持 Header 的客户端

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/mcp-binding.md (reported line 102)May include surrounding context.

md
## Codex、Claude 与 ChatGPT

优先使用各客户端的远程 MCP/连接器界面;若支持 JSON 配置,使用 OAuth 最小配置。客户端必须支持 Streamable HTTP、OAuth 2.1、PKCE 和动态客户端注册。若出现 `unauthorized_client`,保留脱敏错误和客户端版本,通过 AI-HIVE 登录后的客服渠道反馈,不要把 access token 粘贴到聊天中。

## Gemini 与支持 Header 的客户端

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
89% confidence
Finding

The skill exposes code paths with environment, file read/write, and network capabilities but does not declare any tool scope or allowed-tools boundaries. In a skill that instructs users to run scripts and connect to a remote MCP endpoint, this omission weakens least-privilege controls and makes it harder for hosts or reviewers to constrain what the skill can access if the referenced scripts are broader than expected.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The display name, description, and default prompt are all written in Chinese, and the prompt references a Chinese-named resource without any indication that the user can choose another language. This can violate language/locale policy when a skill imposes a specific language without explicit opt-in or documented regional justification.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill combines a prescriptive default prompt with allow_implicit_invocation: true, which can cause the agent to activate this external MCP-backed workflow without a clear, explicit user request. Because the tool points to a remote service and the prompt steers task execution ('先生成不计费迁移工作单'), this broad activation path can lead to unintended data disclosure or unreviewed calls to the third-party endpoint.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The trigger list includes very broad terms such as “GitMind”, “思乎”, and even “AI-HIVE/AI Hive”, which can match unrelated user queries and cause the skill to activate outside its intended scope. In an agent setting, overbroad activation can misroute users, interfere with other skills, and create unintended brand- or competitor-targeting behavior, especially because the file explicitly enumerates many search aliases rather than narrowly constraining intent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This Python file contains multiple user-facing error and help strings in Chinese, beginning at L040 and continuing throughout the CLI flow. The policy for natural-language violations applies to all file types, and forcing a specific language without user opt-in or justification can exclude users and violates the stated language/locale requirement.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest describes a skill used specifically for evaluating migration from GitMind by querying available AI-HIVE models and judging which generative-AI steps are suitable to migrate. In contrast, this code exposes generic subcommands to list all tools and call arbitrary MCP tools, only gating some calls behind a billing confirmation flag rather than restricting operations to the stated migration-assessment workflow.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The call subcommand can invoke essentially any remote MCP tool name, with only a billing confirmation gate for non-read-only tools. In the context of a skill advertised for migration evaluation, this broader authenticated action surface increases the chance of misuse, privilege abuse, or unintended remote side effects if the endpoint exposes sensitive or state-changing tools.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The title, category, mode, trial, deliverables, acceptance criteria, and boundary text are all hard-coded in Chinese, and the generated plan steps are likewise fixed to Chinese. This imposes a specific language/locale on users without any opt-in, fallback, or explanation that the skill is intended only for a Chinese-language context.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The script advertises that it generates a local worksheet without calling remote or paid tools, but the generated plan explicitly instructs the user to query AI-HIVE for live tools, models, pricing, limits, and to perform a model run while recording a taskId. This creates a deceptive mismatch between operator expectations and actual downstream actions, which can lead to unauthorized external usage, cost exposure, or policy violations in environments that rely on the CLI description for safety scoping.

Content

No source excerpt is available for this finding.