Back to skill

Security audit

huawei-cloud-cce-alarm-correlation-engine

Security checks for vulnerabilities and agentic risk

Overview

The skill’s cloud alarm-management capabilities are disclosed, purpose-aligned, and require confirmation before changes.

Use a least-privilege hcloud profile, preferably read-only for inspection. Review every mutation preview, including notification settings, and check dependent alarms before deleting notification rules. Cluster filters do not remove unrelated records from every raw response. Expect Chinese reports and Asia/Shanghai notification defaults. The scanner reported complete coverage; independent review checked executable operations, credentials, temporary files, and confirmation gates, but did not exercise live cloud operations.

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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (12)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

Most dispatch handlers align with the declared purpose. However, huawei_delete_aom_action_rule accepts a region and rule name and forwards a deletion request to aom.delete_aom_action_rule. This is independently available, not limited to deleting an automatically created notification rule during CCE cleanup. Passing a confirmation flag does not remove this undeclared destructive capability. Listing mute rules and resolving a Prometheus instance are consistent supporting operations. The supplied chunk only shows delegation; the underlying AOM implementation is not available for verification.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

Alarm-rule deletion is declared, but notification action rules are a separate resource whose declared operations are limited to query and creation. The code invokes huawei_delete_aom_action_rule without an explicit preview or dry-run argument. Its comment alone does not establish whether deletion is prevented; the CLI implementation is not supplied. Listing mute rules is related supporting functionality and is not independently treated as a mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The primary behavior matches the description. However, notification action-rule deletion is an additional destructive capability: the description only declares querying or creating notification action rules, and its cleanup scope refers to recommended CCE alarm rules. The standalone deletion function can delete a supplied notification rule name, not just an automatically generated rule. Mute-rule querying is also not explicitly declared. Confirmation requirements mitigate accidental changes but do not eliminate the capability mismatch. CCE addon inspection and project resolution are supporting implementation details, not mismatches.

Content

No source excerpt is available for this finding.

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · references/verification-method.md (reported line 53)May include surrounding context.

md
Expected result:

- Cluster-scoped queries only return rules related to the target `cluster_id`.
- `cluster_name` is not accepted by `huawei_list_aom_alarm_rules`.

## Mutation Verification

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · scripts/huawei_cloud/aom.py (reported line 1239)May include surrounding context.

python
if not result.get("success"):
        return None, result.get("error") or "Failed to list alarm rules"
    rule = next((rule for rule in result.get("rules", []) if rule.get("rule_name") == rule_name), None)
    return rule, None


def create_aom_alarm_rule(

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

Notification rules force a Chinese template and Shanghai time zone without user locale selection.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

The notification payload unconditionally selects aom.built-in.template.zh and Asia/Shanghai. The function offers no language or time-zone option, and the file provides no documented regional requirement justifying these enforced locale settings.

Content

No source excerpt is available for this finding.

The skill can delete arbitrary notification action rules, although the manifest declares only querying and creating them.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The manifest distinguishes full alarm-rule management from notification action rules, for which it declares only query and create operations. At L2398-L2403, delete_aom_action_rule accepts any rule name and invokes AOM DeleteActionRule when confirmed, without restricting deletion to notification rules associated with CCE template cleanup. Deleting a shared notification action rule can disrupt alarm delivery beyond the intended cleanup.

Content

No source excerpt is available for this finding.

Alarm reports and analysis output are hard-coded in Chinese without a language choice.

Medium
Category
Output Handling
Confidence
95% confidence
Finding

The report is generated with Chinese headings and descriptions regardless of the user's language, and the function exposes no locale parameter. The analysis path similarly hard-codes Chinese labels, reasons, and report text at L2633–L2644 and L2680–L2684, without a documented language requirement or opt-in.

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/huawei_cloud/common.py (reported line 167)May include surrounding context.

python
command.append("--dryrun")

    try:
        completed = subprocess.run(command, text=True, capture_output=True, timeout=timeout, check=False)
    except subprocess.TimeoutExpired as exc:
        return {
            "success": False,

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/huawei_cloud/common.py (reported line 216)May include surrounding context.

python
if dryrun:
            command.append("--dryrun")
        try:
            completed = subprocess.run(command, text=True, capture_output=True, timeout=timeout, check=False)
        except subprocess.TimeoutExpired as exc:
            return {
                "success": False,

The dispatcher exposes standalone deletion of notification action rules, although the manifest declares only querying and creating them.

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

L273–L281 forward a caller-selected rule name to aom.delete_aom_action_rule, and L390 registers this as an independently callable action. Unlike deletion of automatically created notification rules during CCE template cleanup, this operation is not tied to cleanup and exceeds the manifest's stated query/create scope for notification action rules, potentially removing notification routing.

Content

No source excerpt is available for this finding.