Back to skill

Security audit

huawei-cloud-skill-creator

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with high confidence because scanners report live cloud changes, automatic telemetry, and capabilities beyond its stated skill-generation purpose.

Consider the reported cloud mutations, telemetry, credential handling, and additional installation behavior before installing. The scan is partial: 11 of 13 components were fully inspected, and two scripts had analysis limits, so the results do not establish that the skill is safe.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/test-cli-commands.sh
Finding

Test runner executes cloud mutations without enforcing confirmation or syntax-only mode

Content
View full analysis

Impact

A test invocation can create, modify, or delete cloud resources using the current profile without the promised confirmation or help-only conversion. Impact is limited by the selected commands and the profile's cloud permissions.

Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (21)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
86% confidence
Finding

Quality tooling can support the declared compliance phase, and this helper need not implement the entire workflow. However, it additionally performs remote-code bootstrapping and persistent command installation, capabilities not disclosed by the description's generation and validation claims. The HTTPS host checks and checksum verification constrain the download but do not remove those additional behaviors. The supplied code does not itself demonstrate uploading skill contents or quality reports.

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 208)May include surrounding context.

md
tion-method.md / dataflow-diagram.md / acceptance-criteria.md recommended), scripts/test-cli-commands.sh, scripts/ensure_cli.sh (quality CLI, required — see rul

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 421)May include surrounding context.

md
- **功能测试(Phase 4/5)**:执行 bash scripts/test-cli-commands.sh,传入 -s 参数指向 skill 目录、-e 参数选择执行模式(cli/sdk/api/auto)。

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 465)May include surrounding context.

md
bash scripts/test-cli-commands.sh -s {skill-path} -e {cli|sdk|api} # CLI priority → SDK fallback → API fallback

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 249)May include surrounding context.

md
*Change scope** — A pull request must change only one Skill directory. Use bash scripts/validate-skill.sh -s {skill-path} -b <base-ref> to validate the PR diff

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 259)May include surrounding context.

md
- **Re-validate**: 注入后重跑 bash scripts/validate-skill.sh -s {skill-path},确认 SKILL.md ≤ 500 行、文件 ≤ 30。

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 420)May include surrounding context.

md
- **结构/规范校验(Phase 3/6)**:执行 bash scripts/validate-skill.sh 并传入 -s 参数指向目标 skill 目录(如 -s skills/devtools/common/huawei-cloud-skill-creator)。

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 459)May include surrounding context.

md
bash scripts/validate-skill.sh -s {skill-path} # Check against 华为云Skill检查规范

Resource lifecycle execution contradicts the explicit packaging-only, no-cloud-mutation guarantee.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

The overview promises packaging-only behavior, but mandatory lifecycle testing explicitly creates, modifies, and destroys real cloud resources. Per-step confirmation reduces risk, but the contradiction can still mislead users about operational scope and expose production resources to changes, costs, or deletion.

Content

No source excerpt is available for this finding.

Test commands execute live without the documented mutation safeguards.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

Commands from templates/test-vars.json are executed directly after an executable-name allowlist check, without classifying cloud operations or obtaining mutation approval. Despite the header's promise of help-only checks for mutations, hcloud write operations and mutating curl requests can run live using available authority; argv-safe execution prevents shell injection but does not make those operations safe.

Content

No source excerpt is available for this finding.

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

Generated documentation is required to use English without user language opt-in.

Medium
Category
Output Handling
Confidence
99% confidence
Finding

L203 mandates English for the generated SKILL.md description, headings, examples, and explanatory content. Allowing supplementary Chinese documentation does not provide a choice for the main document, and no justification for the English-only requirement is given.

Content

No source excerpt is available for this finding.

Mandatory telemetry installation and propagation exceed the skill-generation purpose.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The skill mandates installation of a telemetry intermediary, routes every hcloud invocation through it, and propagates this requirement into every generated skill. This extends the trust boundary around credential-bearing processes without defining telemetry fields, destinations, or redaction guarantees; the opt-out limits reporting but does not remove the mandatory intermediary.

Content

No source excerpt is available for this finding.

Test execution requires reading credential values despite an explicit prohibition on doing so.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The skill prohibits reading AK/SK values but subsequently directs the agent to scan matching environment-variable key-value pairs. Authentication libraries legitimately consume credentials, but broad agent-directed inspection unnecessarily exposes secrets to tool output, conversation context, or diagnostic logs and may select unintended credentials.

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 482)May include surrounding context.

md
Fact-based** — endpoints only from SDK _http_info/API Explorer (no inference); commands per Phase 2 conclusions; no endpoint → mark ⛔
- **Phase 4-6 Verify** — every command really executed (mutating ops need confirmation); Phase 6: cleanup + compliance + completeness; fix failures, re-verify
- **Credential Security** — no hardcoded AK/SK; read from env vars (HUAWEI_*/HW_*/HWC_* AK/SK markers) or active CLI profile; never hcloud configure set with literal secrets
- **Credentials Mandatory** — if AK/SK missing after the Pre-check, output the env-var setup template for out-of-band fill; never ask the user to paste AK/SK into chat; if still unconfigured, terminate
- **BSS SDK** must use GlobalCredentials + with_endpoints, not BasicCredentials with with_region.
- **BSS 区域**:BSS 服务仅支持 cn-north-1 等区域;涉及 BSS 的示例命令应标注 --cli-region=cn-north-1(SDK 初始化也用 cn-north-1),test-defaults.json 中 region: cn-north-1 即为此口径。
- **Cleanup** — resources created during lifecycle testing must be released in Phase 6

Live cloud command execution and resource lifecycle testing lack an impact warning.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The readiness checklist requires executing all CLI commands, testing resource lifecycles, and cleaning up test resources; these are workflow requirements rather than a tool catalog. The file provides no warning that these steps can change cloud resources, incur charges, or delete resources during cleanup.

Content

No source excerpt is available for this finding.

The guide prescribes automatic, opt-out telemetry reporting that is not justified by the skill's declared creation and validation workflow.

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

The guide explicitly states that telemetry reporting is automatic unless disabled. Although quality validation supports the manifest's compliance-checking purpose, automatic telemetry reporting is a separate capability not declared in that purpose; this file does not establish what information is reported or where it goes.

Content

No source excerpt is available for this finding.

The workflow includes cloud-resource lifecycle testing and cleanup without warning about resource modification or deletion.

Medium
Category
Output Handling
Confidence
86% confidence
Finding

Phases 5 and 6 place resource lifecycle testing and resource cleanup in the skill's own execution pipeline, rather than merely listing available tools. The file provides no warning that these steps can change or delete cloud resources, nor does it specify a safe cleanup scope.

Content

No source excerpt is available for this finding.

The creator's required IAM policy requests unrelated credential and billing-coupon permissions (L17–L19).

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

The policy labels credential-listing and billing-coupon permissions as generally required for skill creation, although neither is inherently necessary to generate skills or inspect CLI/SDK availability. If accepted and effective, these grants expand access to sensitive IAM or billing metadata beyond the requested task; the validity of the exact action names is not established here.

Content

No source excerpt is available for this finding.

The resource lifecycle verification instructs resource creation and destruction without a safety warning.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

Phase 5 directs a Create → Verify → Query → Destroy lifecycle test without warning that cloud resource creation may incur charges and destruction may permanently remove resources or data. It also does not restrict the test to disposable resources, leaving existing user resources unprotected by an explicit scope constraint.

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 · scripts/validate-skill.sh (reported line 117)May include surrounding context.

sh
# ak=..., sk=...) inside a BasicCredentials(...) call — both are
    # in-band secret-entry forms that put AK/SK on a command
    # line / SDK kwarg, contradicting the Pre-check contract:
    #   'NEVER ask the user to type or paste AK/SK in chat; user must
    #    set env vars in shell profile out-of-band and re-run.'
    # Whitelist: lines that contain the patterns inside NEVER / Do NOT
    # / 禁止 / forbidden contexts (prose warnings about the anti-pattern