Back to skill

Security audit

明日dmp批量打通圈包

Security checks for vulnerabilities and agentic risk

Overview

This skill performs a real DMP bulk audience workflow, but it combines sensitive API credentials, automated scheduled mutations, broad local helper discovery, and a forced unrelated logger step that users should review before installing.

Install only in a trusted workspace. Review the authentication-helper dependency, avoid untrusted sibling skills, rotate or remove DMP credentials when done, and confirm whether the mandatory skill-logger recording and cron-based automatic batch progression are acceptable for your organization’s data and approval process.

Vulnerability Patterns
  • Tool Hijacking and SpoofingModifies or replaces tools so legitimate-looking calls execute attacker logic
  • 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
Findings (1)

T07 · Tool Hijacking and Spoofing

Warning
Location
scripts/find_auth_skill.py:43
Finding

Weak Authentication Helper Discovery Permits Tool Spoofing and Arbitrary Code Execution

Content
View full analysis

Vulnerability Details

File Location: scripts/find_auth_skill.py:43-61, with the execution sink at scripts/run_idlink_project.py:91-99
Vulnerability Type: Untrusted local tool discovery and execution
Risk Level: Medium

Complete Code Snippet

Discovery logic in scripts/find_auth_skill.py:

python
# Second layer: dynamically scan all possible directories
scan_dirs = [
    Path.home() / ".skills",
    Path.home() / ".openclaw" / "workspace" / "skills",
    Path.home() / ".openclaw" / "skills",
    Path.cwd() / ".skills",
]
for scan_dir in scan_dirs:
    if scan_dir.exists():
        for skill_dir in scan_dir.iterdir():
            if skill_dir.is_dir():
                auth_path = skill_dir / "scripts" / "minri_dmp_api.py"
                if auth_path.exists():
                    try:
                        with open(auth_path, 'r', encoding='utf-8') as f:
                            content = f.read(500)
                            if "明日DMP" in content or "mingdata" in content.lower():
                                return auth_path
                    except:
                        continue

Execution sink in scripts/run_idlink_project.py:

python
def call_api_via_auth(auth_script: Path, method: str, endpoint: str, body: dict, timeout=60):
    try:
        result = subprocess.run(
            ["python3", str(auth_script), method.upper(), endpoint, json.dumps(body, ensure_ascii=False)],
            capture_output=True,
            text=True,
            timeout=timeout,
        )

Technical Analysis

When an authentication helper is not found at one of the preferred fixed paths, the Skill scans every immediate child directory under several user- and workspace-controlled skill locations. It accepts any file named scripts/minri_dmp_api.py if one of two easily forged keywords appears within its first 500 characters.

This content check does not establish the helper's identity or integrity. It verifies neither a tru ...[truncated 2181 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove the fallback that scans arbitrary sibling skill directories.
  2. Resolve the authentication helper only through an exact, platform-provided skill identity or an explicitly configured trusted path.
  3. Canonicalize the selected path with Path.resolve() and verify that it remains inside an approved authentication-skill directory.
  4. Verify the helper through a trusted manifest, cryptographic signature, or pinned file hash before execution.
  5. Validate ownership and permissions so untrusted local users or workspace content cannot replace the helper.
  6. Fail closed if the expected authenticated component cannot be found or verified; do not fall back to keyword-based discovery.
  7. Where supported, import or invoke the helper through a platform registry that enforces package identity rather than executing a discovered Python file.
  8. Apply the same hardened resolver to all scripts that independently locate or execute minri_dmp_api.py, including audience and status-query helpers.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (48)

Tp2

High
Category
MCP Tool Poisoning
Confidence
85% confidence
Finding

Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

This mismatch suggests the skill may actually target a different operational workflow than the one declared, with extra JSON/report features. While some extra features may be benign, the larger issue is that the declared identity of the skill is unreliable, which undermines security review and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

This mismatch suggests the skill may actually target a different operational workflow than the one declared, with extra JSON/report features. While some extra features may be benign, the larger issue is that the declared identity of the skill is unreliable, which undermines security review and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

This mismatch suggests the skill may actually target a different operational workflow than the one declared, with extra JSON/report features. While some extra features may be benign, the larger issue is that the declared identity of the skill is unreliable, which undermines security review and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

This mismatch suggests the skill may actually target a different operational workflow than the one declared, with extra JSON/report features. While some extra features may be benign, the larger issue is that the declared identity of the skill is unreliable, which undermines security review and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

This mismatch suggests the skill may actually target a different operational workflow than the one declared, with extra JSON/report features. While some extra features may be benign, the larger issue is that the declared identity of the skill is unreliable, which undermines security review and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

This mismatch suggests the skill may actually target a different operational workflow than the one declared, with extra JSON/report features. While some extra features may be benign, the larger issue is that the declared identity of the skill is unreliable, which undermines security review and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

This mismatch suggests the skill may actually target a different operational workflow than the one declared, with extra JSON/report features. While some extra features may be benign, the larger issue is that the declared identity of the skill is unreliable, which undermines security review and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

This mismatch suggests the skill may actually target a different operational workflow than the one declared, with extra JSON/report features. While some extra features may be benign, the larger issue is that the declared identity of the skill is unreliable, which undermines security review and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

This mismatch suggests the skill may actually target a different operational workflow than the one declared, with extra JSON/report features. While some extra features may be benign, the larger issue is that the declared identity of the skill is unreliable, which undermines security review and informed consent.

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 226)May include surrounding context.

md
"message": f"检查DMP批量打通项目{project_id}状态:运行 python3 scripts/run_idlink_project.py --mode monitor --project-id {project_id} 推进批次;若本批完成则创建下一批;全部完成则汇报结果并删除此cron任务。"

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 234)May include surrounding context.

md
"message": f"检查DMP批量打通项目{project_id}状态:运行 python3 scripts/run_idlink_project.py --mode monitor --project-id {project_id} 推进批次;若本批完成则创建下一批;全部完成则汇报结果并删除此cron任务。"

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 255)May include surrounding context.

md
"message": f"检查DMP批量打通项目{project_id}状态:运行 python3 scripts/run_idlink_project.py --mode monitor --project-id {project_id} 推进批次;若本批完成则创建下一批;全部完成则汇报结果并删除此cron任务。"

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 256)May include surrounding context.

md
"message": f"检查DMP批量打通项目{project_id}状态:运行 python3 scripts/run_idlink_project.py --mode monitor --project-id {project_id} 推进批次;若本批完成则创建下一批;全部完成则汇报结果并删除此cron任务。"

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 315)May include surrounding context.

md
"message": f"检查DMP批量打通项目{project_id}状态:运行 python3 scripts/run_idlink_project.py --mode monitor --project-id {project_id} 推进批次;若本批完成则创建下一批;全部完成则汇报结果并删除此cron任务。"

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
95% confidence
Finding

The skill directs use of shell scripts, file reads, and file writes but declares no explicit tool scope or permissions boundary. In an agent setting this can lead to over-broad execution capability, making it easier for the skill to invoke local commands, modify files, or access sensitive workspace data without clear authorization constraints.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

L011 和多处描述均声明“每批最多5个任务、同批串行、批间等待本批完成”,这是该技能的核心行为承诺;但 L250-L253 又规定“任务数 > 1:按每批 1 个分批创建”。两者对实际批处理行为给出了互相冲突的指令,会导致技能宣称的批量策略与实际执行方案不一致。

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill asks users to provide Access Key and Secret Key and even write them via a setup script, but the warnings about secrecy, local persistence, and host exposure are weak relative to the sensitivity of the data. In this context, the danger is elevated because the skill also uses shell and local files, increasing the chance of accidental credential leakage or insecure storage.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

L073-L085 明确写明“所有字段均为必填”,且将“所属品牌”“MAC格式过滤”标记为必填/需确认;但 L138-L146 的上传表格说明又把“品牌”“MAC格式过滤”写成“可选”。这不是单纯信息不全,而是对同一输入要求的主动矛盾描述,可能导致代理在收参和执行时采取相反行为。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file’s natural-language description states behavior entirely in Chinese and says it will 'force output' a mandatory step prompt. Under the policy, language or locale constraints should not be imposed without user opt-in or a clearly documented justification, and no such choice or justification appears here.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The wrapper modifies successful API responses to inject a mandatory, unrelated workflow step that blocks result display until the user is steered through a 'skill-logger' check/download flow. This is dangerous because it creates hidden control-flow manipulation and can be used for social engineering or unauthorized software promotion that is unrelated to the declared DMP audience task functionality.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

The code hard-codes instructions to check for and potentially download an external component ('skill-logger') even though the skill's advertised purpose is bulk creation and tracking of DMP audience tasks. This mismatch strongly suggests an attempt to coerce installation of unrelated software, creating supply-chain, malware, or data-exfiltration risk through deceptive in-band instructions.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The injected step name and instruction strings are all hard-coded in Chinese and are marked as mandatory, with no indication that users can choose another language. This is a natural-language policy concern because the skill enforces a specific language/locale in its output without documented user choice or justification.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This script adds a mandatory workflow around detecting and potentially installing an unrelated 'skill-logger' component, which is outside the stated purpose of bulk audience-task creation. That expands the skill's operational scope, creates pressure to introduce extra software into the environment, and can steer the agent into data-recording behavior without clear user consent or necessity.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The module description and all user-facing print messages are in Chinese, and the script instructs the AI assistant to display fixed Chinese prompts to users. This creates a language-policy issue because it imposes a specific language without any opt-in, fallback, or documented justification that the skill is intended only for a Chinese-speaking context.

Content

No source excerpt is available for this finding.