Back to skill

Security audit

huawei-cloud-sac-intl-find-and-deploy

Security checks for vulnerabilities and agentic risk

Overview

Cloud deployments require your approval, but automatic installation of extra software and unclear telemetry handling need review.

Before installing, ask for optional installation of the quality CLI and a clear explanation of its telemetry fields and destination. Disabling reporting does not prevent its installer from running. Keep TLS verification enabled, use scoped temporary credentials through environment variables, and review the selected template and Terraform plan. All 14 bundled files were inspected, including the scanner's coverage gaps; downloaded CLI packages and live templates remain outside this review. No malicious behavior or credential exfiltration was established.

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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (17)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied script's primary purpose is installing a skill-quality tool, not finding or deploying SAC solutions. Neither that tool nor its installation is disclosed in the declared workflow, and the code shows no connection establishing it as a necessary supporting implementation detail. Manual invocation, checksum validation, and archive checks do not resolve this purpose mismatch. The script does not execute the installed CLI or visibly access cloud credentials; the downloaded tool's subsequent behavior cannot be determined from this chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied chunk primarily bootstraps a skill-quality tool, not a SAC search or Terraform deployment dependency identified in the description. It does not demonstrate any connection between this quality CLI and the declared workflow, so its network access and persistent executable installation constitute an undeclared capability rather than an evident supporting implementation detail. Checksum validation and archive screening reduce installation risks but do not resolve the purpose mismatch. The chunk does not establish credential collection, telemetry transmission, or malicious behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The supplied chunk concerns skill-quality-cli management rather than SAC catalog lookup or Terraform deployment. Although dependency installation can be a supporting implementation detail, no connection between this quality CLI and the declared workflow is shown. The telemetry setting only documents how to disable reporting; it does not establish that this chunk transmits data. The referenced installer scripts are not supplied, so their internal behavior cannot be assessed.

Content

No source excerpt is available for this finding.

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · scripts/install_cli.sh (reported line 28)May include surrounding context.

sh
API_URL="https://skillsapi.developer.myhuaweicloud.com/api/quality/cli/latest"

manifest="$(curl -fsSL --connect-timeout 5 --max-time 20 "${API_URL}")"
v="$(printf '%s' "$manifest" | python3 -c 'import sys,json;print(json.load(sys.stdin)["version"])')"
case "$(uname -s)" in
    Linux) ;;

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.

Automatic quality telemetry and installation of its CLI are unrelated to the declared catalog-search and Terraform-deployment capabilities.

Medium
Category
Excessive Agency
Confidence
92% confidence
Finding

The skill directs automatic installation of an auxiliary executable and default-on telemetry without a distinct approval gate for either activity. This expands a cloud deployment session's trust boundary to unnecessary third-party tooling; a compromised dependency could run with the agent's filesystem and cloud-credential access, although this source does not establish actual credential transmission.

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 40)May include surrounding context.

md
### Security

- 🚫 Never expose AK/SK/Token in conversation or output
- 🚫 Never ask user to paste AK/SK/Token in chat; prefer env vars or manual tfvars edit
- 🚫 Never read or display `terraform.auto.tfvars.json` in conversation context (contains credentials)
- ✅ Only temporary credentials with SecurityToken, least privilege, shortest TTL
- ✅ Modification ops (`apply`, `destroy`) require explicit user confirmation

Unbounded Output

Medium
Category
Output Handling
Confidence
60% confidence
Finding

Output size or generation rate is not bounded. Unbounded output enables denial-of-service through resource exhaustion, log flooding, or context-window stuffing.

Content

Scanner excerpt · SKILL.md (reported line 127)May include surrounding context.

md
#### 3. Stage B — pick solution, then show ALL regions

Wait for the user to pick exactly one solution+variant. Then run Core Commands step 1b: it lists **every** region for that variant (no truncation, no hidden cuts). Present rows as `solution_name | variant | region_label (console_region)` with `template:` and `deploy:` lines. Wait for the user to explicitly pick one region. Never auto-pick. Mention the `deploy_link` supports console self-service without credentials.

### Part 2 — Deploy

The installation guide introduces an unrelated telemetry CLI outside the skill's declared catalog-search and Terraform-deployment purpose.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

L21–L27 instruct users to run installation scripts and manage skill-quality-cli for telemetry. This auxiliary telemetry capability is not declared in the manifest and is not an obvious requirement for searching SAC solutions or deploying their Terraform templates; the provided file does not establish what data it reports.

Content

No source excerpt is available for this finding.

The guide recommends `--insecure` without warning that it disables TLS certificate verification.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

The guide recommends disabling TLS certificate verification for catalog and Terraform-template downloads without explaining the loss of server authentication. An attacker able to intercept those connections could substitute catalog entries or executable Terraform configuration, potentially affecting the local agent and cloud account.

Content

No source excerpt is available for this finding.

Automatically installs an unrelated skill-quality CLI outside the SAC deployment workflow.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

When the CLI is absent, this readiness helper automatically retrieves a latest-release manifest and installs executable components into the user's binary directory without an installation-specific consent check. TLS, mandatory checksums, and archive checks reduce tampering risks, but the checksum comes from the same manifest that selects the package, so a compromised release source can still supply persistent executable code.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · scripts/ensure_cli.sh (reported line 89)May include surrounding context.

sh
echo "警告: 解压失败或 tar 包含不安全条目, 拒绝安装" >&2
        return 1
    fi
    mkdir -p ~/.local/bin/skill-quality-cli.d || { echo "警告: 创建安装目录失败" >&2; return 1; }
    [ -f "${tmp}/skill-quality-cli" ] || { echo "警告: 安装包内容缺失(skill-quality-cli)" >&2; return 1; }
    # 主二进制: 复制/可执行位失败必须显式失败(不再静默吞错)
    cp "${tmp}/skill-quality-cli" ~/.local/bin/skill-quality-cli || { echo "警告: 复制 skill-quality-cli 失败" >&2; return 1; }

L26–L88 download and persistently install an unrelated skill-quality CLI outside the SAC deployment workflow.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The manifest describes SAC catalog discovery and Terraform-based solution deployment, not installation of skill-quality tooling. This script retrieves a separate CLI package from a quality-service endpoint (L26–L48) and installs executable files and supporting modules under ~/.local/bin (L78–L88). Manual invocation and checksum verification do not establish a need for this additional capability within the declared workflow.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · scripts/install_cli.sh (reported line 78)May include surrounding context.

sh
echo "错误: 解压失败或 tar 包含不安全条目, 拒绝安装" >&2
        exit 1
    fi
mkdir -p ~/.local/bin/skill-quality-cli.d
cp "${tmp}/skill-quality-cli" ~/.local/bin/skill-quality-cli || { echo "错误: 复制 skill-quality-cli 失败" >&2; exit 1; }
chmod +x ~/.local/bin/skill-quality-cli || { echo "错误: 设置可执行位失败" >&2; exit 1; }
if [ -f "${tmp}/skill-quality-cli.bin" ]; then

Credentials can be supplied through command-line arguments despite the manifest specifying environment-variable credential input.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The parser permits access keys, secret keys, and security tokens as command-line arguments, despite the workflow's environment-based credential guidance. When used, these options can expose credentials through shell history, process inspection where permitted, or agent command logs; temporary credentials reduce the exposure window but remain usable until expiration.

Content

No source excerpt is available for this finding.

Mandatory catalog announcements prescribe Chinese output without offering a language choice.

Low
Category
Output Handling
Confidence
86% confidence
Finding

The mandatory freshness announcement instructs the agent to always begin results with a Chinese-language sentence, and L221 reinforces that wording as a verification requirement. No user language choice or justification for requiring Chinese is provided.

Content

No source excerpt is available for this finding.

The documented exact solution selector performs substring matching instead.

Low
Category
Excessive Agency
Confidence
98% confidence
Finding

The documented exact solution selector uses substring matching, so selecting one solution can also return differently named solutions containing the same text. In this deployment workflow, downstream automation that assumes the selection is unique could choose the wrong Terraform template, potentially provisioning unintended resources. This script does not itself deploy anything, and the stated solution, variant, region, and deployment confirmations substantially reduce the risk.

Content

No source excerpt is available for this finding.