Back to skill

Security audit

huawei-cloud-cloudrobo-data-pipeline

Security checks for vulnerabilities and agentic risk

Overview

This skill coherently supports cloud data pipelines, with disclosed credential use and explicit confirmation for consequential actions.

Use credentials limited to the intended workspace and storage paths. Review selected data, operators, resources, and costs before approving execution or a Web retry, and keep TLS verification enabled. Stopping monitoring or terminating orchestration does not stop existing cloud jobs. External CLI/SDK packages and the companion skill were outside this artifact review.

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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (69)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The retry interface lets users change task names, resource pools, worker specifications, environment variables, and input mounts, then submits those settings to the run's /actions endpoint. Its confirmation and success messages explicitly describe creating a new cloud task. This exceeds passive monitoring and contradicts the declared CLI-only submission model, even though subsequent orchestration remains assigned to the terminal Agent. The supplied chunk shows the frontend submission request, not the backend implementation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared purpose includes selecting and composing algorithm assets into pipelines, but does not include releasing asset versions. Updating an asset's release status is a separate state-changing capability, not merely a supporting detail of pipeline drafting, task submission, or monitoring.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Most of this module supports the declared monitoring functionality. However, retry_draft() and action() implement more than monitoring: users can edit permitted task settings, confirm a retry, and have the web backend create a new cloud task and replace the tracked task ID. This is a material submission capability not disclosed by the description's CLI-only submission model. Confirmation, validation, and concurrency safeguards do not remove that mismatch. The absence of YAML-generation code in this particular monitoring module is not itself evidence of a mismatch.

Content

No source excerpt is available for this finding.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 28)May include surrounding context.

md
[User Guide 用户指南](references/user-guide.md).

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 1291)May include surrounding context.

md
- [User Guide 用户指南](references/user-guide.md) — Complete user-facing workflow, features, examples and troubleshooting

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 149)May include surrounding context.

md
[CLI Installation Guide](references/cli-installation-guide.md)

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 168)May include surrounding context.

md
[CLI Installation Guide](references/cli-installation-guide.md#environment-preflight).

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 1287)May include surrounding context.

md
- [CLI Installation Guide](references/cli-installation-guide.md) — CloudRobo CLI installation and configuration

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 1306)May include surrounding context.

md
sk <sk>` or `HUAWEI_CLOUD_AK` / `HUAWEI_CLOUD_SK` (see [CLI Installation Guide](references/cli-installation-guide.md)) |

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 1295)May include surrounding context.

md
- [Verification Method](references/verification-method.md) — Verification methods and CLI command reference

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 1357)May include surrounding context.

md
See [Verification Method](references/verification-method.md) and

The Web monitor directly submits replacement cloud tasks despite the manifest's CLI-only Agent task-submission restriction.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

The retry action directly invokes self.client.create_task from the Web backend, contradicting the advertised CLI-only Agent submission boundary. Authentication, explicit confirmation, configuration validation, and state locking reduce unauthorized-use risks, but this alternative submission path still bypasses the required Agent workflow and can create billable cloud workloads.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

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

Content

Scanner excerpt · scripts/pipeline-monitor/tests/test_monitor.py (reported line 2572)May include surrounding context.

python
self.assertEqual(list(responses), [200] * len(scripts))
            self.assertEqual(request('GET', '/unknown.js', cookie)[0], 404)
            self.assertEqual(request('GET', '/server.py', cookie)[0], 404)
            self.assertEqual(request('GET', '/../../etc/passwd', cookie)[0], 404)
        finally:
            server.shutdown(); server.server_close(); thread.join()

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · scripts/platform_utils.py (reported line 137)May include surrounding context.

python
def utf8_environment() -> dict:
    return {**os.environ, 'PYTHONIOENCODING': 'utf-8', 'PYTHONUTF8': '1'}


def configure_console() -> None:

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, file_read, file_write, network, shell.

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.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% 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.md (reported line 244)May include surrounding context.

md
When the user wants to create/design/configure a pipeline without YAML, or revise an
existing pipeline, read [Guided Custom Pipeline Authoring](references/custom-pipeline-authoring.md).
Reuse information already given; guide missing choices from data stage/source and desired
output through steps/operators, optional params, resources and failure policy. Do not ask
the user to write YAML or silently choose the full built-in template.

If the user only says “创建管线” or “引导我”, start with “希望处理什么数据、得到什么结果?

The standalone phrase “引导我” (“guide me”) activates pipeline authoring without requiring pipeline context.

Medium
Category
Trigger Abuse
Confidence
94% confidence
Finding

The instruction explicitly starts the pipeline interview when the user says only “引导我,” a common request that could refer to any task. Unlike the domain-specific pipeline triggers, this condition does not require an established CloudRobo or data-pipeline context.

Content

No source excerpt is available for this finding.

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.md (reported line 421)May include surrounding context.

md
Say “执行计划已确认,运行记录已创建,准备执行。是否打开本次运行监控?终端会继续跟踪。”
Local init does not establish successful cloud submission; the page can already show the
confirmed plan and pending/not-applicable input preparation without a task ID.
If the user already requested monitoring, open it now without asking again. Otherwise
offer “打开本次运行监控 / 仅终端跟踪”; do not ask demo vs real mode. When opening is
accepted, start/reuse the service and open this run before proceeding to Stage 1. Keep terminal
orchestration moving while awaiting this optional preference; do not set await-decision

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 · references/iam-policies.md (reported line 91)May include surrounding context.

md
| Type | Operations | Rule |
|------|------------|------|
| Read-only | `workspace current/list`, `asset list-publication-assets/list-assets/show-asset/list-versions/show-version`, `dataset proc\|eval list-tasks/show-task/wait-task/get-log/download-log`, `dataset eval get-preview`, `resource list-pools` | May run without confirmation |
| Mutating | `asset create-asset`, `asset create-version`, `asset import-asset`, `asset update-version`, `dataset proc create-task`, `dataset eval create-task` | Only after the user confirms the execution plan (SKILL.md "Execution Plan Confirmation") |
| Local config | `config set`, `workspace use` | Run by the user (credentials are never handled by the agent) |

Unsafe Defaults

Medium
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool defaults are unsafe or overly permissive (e.g. disabled TLS verification, no authentication, world-writable permissions). Unsafe defaults widen the attack surface.

Content

Scanner excerpt · references/iam-policies.md (reported line 125)May include surrounding context.

md
- **Confirmation gate** — every mutating operation runs only after plan confirmation
- **`obs://` only** — no other object-storage schemes or public URLs as inputs
- **Report links** — `get-preview` returns temporary OBS links; do not publish them
- **SSL verification** — can be disabled for debugging (`CLOUDROBO_VERIFY_SSL=false`) but
  should be enabled in production
- **Traffic logging** — `CLOUDROBO_LOG_TRAFFIC=true` enables request/response logging for
  debugging; disable in production to avoid credential leakage in logs

Unbounded Resource Access

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill allows unbounded resource consumption (API calls, storage, compute). Without rate limits or quotas, a compromised or misbehaving agent can cause denial-of-service or cost overruns.

Content

Scanner excerpt · references/web-monitor.md (reported line 575)May include surrounding context.

md
浏览器展示的云端字符串经过转义;预览链接只允许 http/https,新窗口使用 noopener/noreferrer。
本地文件、日志、task_id 均做运行归属或列表校验。服务仅调用上述 SDK 方法,最多四个并行操作,每次操作独享 HTTP 会话并在结束时关闭。
连接超时 10 秒,单次读取等待上限为查询 50 秒、创建 110 秒(不是整个复合 SDK 操作的总时限)。
传输 max_retries=0,禁用连接/超时等瞬时故障的自动重试;SDK 自身的鉴权签名修正和标准重定向行为保留。
状态轮询失败仍按退避周期再查,预览失败保留手动重试,创建结果不明不自动再提交。
创建已发出后遇到超时、网络中断、5xx 或缺失有效任务 ID,保留 pending_action,由智能体通过 CLI 核实。
明确的 4xx 拒绝(408 除外)或发出前的配置/字段校验失败可清理提交锁。关闭 SDK 流量日志,不记录签名头或凭据。

Unbounded Resource Access

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill allows unbounded resource consumption (API calls, storage, compute). Without rate limits or quotas, a compromised or misbehaving agent can cause denial-of-service or cost overruns.

Content

Scanner excerpt · scripts/pipeline-monitor/sdk_adapter.py (reported line 145)May include surrounding context.

python
self.response_status = response.status_code
            return response

    transport = WebHttpClient(config=Config(), max_retries=0)
    transport._log_traffic = False  # Never log signed requests or credentials.
    try:
        clients = {'dataset': DatasetClient(transport), 'asset': AssetClient(transport)}

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/pipeline-monitor/server.py (reported line 444)May include surrounding context.

python
if args.open:
            command += ['--open']
        with tempfile.TemporaryFile() as errors:
            process = subprocess.Popen(command, stdin=subprocess.DEVNULL, stdout=subprocess.DEVNULL,
                                       stderr=errors, env=utf8_environment(), **detached_process_options())
            for _ in range(300):
                time.sleep(.1)

The Web monitor implements cloud-task resubmission despite the manifest's CLI-only Agent task-submission boundary.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The frontend collects retry configuration and posts an explicitly confirmed retry action to the Web API, exposing a task-submission workflow outside the advertised CLI-only Agent boundary. This is not an independent client-side authorization bypass, and the confirmation dialog mitigates accidental submission, but the interface enables the noncompliant cloud-task creation path.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/pipeline-monitor/tests/test_monitor.py (reported line 441)May include surrounding context.

python
assert platform_utils.IS_WINDOWS
'''
        env = {**platform_utils.utf8_environment(), 'PYTHONPATH': os.pathsep.join((str(MONITOR), str(MONITOR.parent)))}
        process = subprocess.run([sys.executable, '-c', code], env=env, capture_output=True,
                                 text=True, encoding='utf-8', timeout=10)
        self.assertEqual(process.returncode, 0, process.stderr)