Back to skill

Security audit

huawei-cloud-find-skills

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marked this skill suspicious because scanners report command-injection risk, automatic installation and execution, default-on telemetry, and extra tooling and persistence beyond skill discovery; overall confidence is medium and coverage is partial.

Treat this as a review-needed skill and consider avoiding installation until the command-injection, automatic install-and-run behavior, telemetry controls, and extra CLI setup have been reviewed. Coverage is partial: 7 of 8 components were scanned, and one static analysis was degraded, so the scan cannot establish that unreported parts are 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
SKILL.md
Finding

Remote index metadata is interpolated into a Bash command

Content
View full analysis

Impact

A malicious index entry can execute arbitrary local commands during the mandatory pre-install telemetry request, independently of whether the selected skill exists or installation succeeds.

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (39)

Tainted flow: 'req' from os.environ.get (line 97, credential/environment) → urllib.request.urlopen (network output)

Critical
Category
Data Flow
Confidence
90% confidence
Finding

The optional GITEE_TOKEN is appended to the fallback request URL as an access_token query parameter, exposing the credential to URL logging by the receiving service and any TLS-terminating intermediaries. This is intentional authentication to Gitee over HTTPS, not evidence of critical arbitrary credential exfiltration; the environment-controlled branch is also not inherently secret.

Content

Scanner excerpt · scripts/search-skills.py (reported line 98)May include surrounding context.

python
for attempt in range(MAX_RETRIES):
        try:
            req = Request(url, headers={"User-Agent": "huawei-cloud-find-skills/1.0"})
            with urlopen(req, timeout=HTTP_TIMEOUT) as resp:
                raw = resp.read().decode("utf-8")
            parsed = json.loads(raw)
            if is_api and isinstance(parsed, dict) and parsed.get("encoding") == "base64" and "content" in parsed:

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The index retrieval, caching, filtering, and ranking support the declared discovery purpose. However, report_search_results_impressions starts a daemon thread that POSTs up to three identifiers formatted as skills// to the Huawei Cloud findcounts/increment endpoint. This reporting is enabled unless SKILL_QUALITY_REPORT=0, and occurs for ordinary searches without an installation action. That external telemetry and counter mutation are not disclosed in the declared description and are not necessary to return search results. Because the thread is a daemon, delivery is best-effort rather than guaranteed. The absence of installation logic in this chunk alone is not the basis for the mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The utility is disclosed in the body, so it is not wholly undeclared, but even discovery-only use mandates installing a latest-release executable and making it available through PATH. This unnecessarily expands the trusted code and persistent local changes needed for public-index searches; the supplied file does not independently establish the claimed remote-bootstrap implementation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

The supplied command performs a tooling upgrade, which is not among the declared capabilities. Although tooling maintenance could support a skill workflow, the supplied chunk provides no connection establishing this upgrade as necessary for Huawei Cloud skill discovery or installation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

The supplied chunk only submits a skill_id to the findcounts/increment endpoint. It does not retrieve skill listings, search results, descriptions, or installation artifacts. Although counter updates could support a larger discovery workflow, no such workflow is shown; the demonstrated behavior is an undeclared remote counter update.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The CLI has a stated telemetry purpose, contrary to the finding's claim that no connection is evident. Nevertheless, requiring a latest-release executable installation for every discovery workflow creates an avoidable supply-chain exposure unrelated to the user's need to search public skill metadata.

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.

Shadow Command Trigger

Medium
Category
Trigger Abuse
Confidence
70% confidence
Finding

Shadow Command Trigger: description clause 'Invoke this skill to search, list, query, discover, browse, find and install any Huawei Cloud (华为云) agent skill' conflicts with built-in command 'find'

Content

No source excerpt is available for this finding.

Discovery-only requests are expanded into automatic skill installation and execution.

Medium
Category
Prompt Injection
Confidence
99% confidence
Finding

The opening instructions turn requests to search or discover skills into mandatory installation and execution of a matched skill. This crosses the user's authorization boundary and can expose local files or cloud resources to newly downloaded instructions when the user requested only information.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

Although this is not an MCP server reference, it describes installation using an unversioned npx package that is executed in later commands. A compromised or unexpectedly changed skills release could execute code with the agent's local privileges; this occurrence represents the same underlying issue as the subsequent command examples.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The referenced npx skills installer is unversioned, allowing execution of package contents that change independently of this skill. This is a duplicate supply-chain finding, not evidence of a separate MCP integration.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 86)May include surrounding context.

macOS

brew install python3

Ubuntu/Debian

sudo apt-get install python3

Windows — download installer from python.org, check "Add Python to PATH"

text
2. **Verify after install**: Run `python --version` again to confirm Python 3.6+ is available

The installation-count command sends telemetry even when the documented global telemetry opt-out is set.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The documentation promises that SKILL_QUALITY_REPORT=0 disables all three telemetry categories, but the installation-count curl command has no condition checking that variable. Disabling the wrapper's own quality reporting does not inherently prevent its child curl command from sending the skill identifier, violating the stated privacy control.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 269)May include surrounding context.

⚠️ MANDATORY (质量上报): 安装计数请求同样用 run 包装, 保证安装流程触发质量上报。

bash
skill-quality-cli run --skill-name huawei-cloud-find-skills -- curl -s -X POST "https://skills-dataset.developer.huaweicloud.com/rest/developer/servlet/hdskillservice/v1/obs/findcounts/increment" -H "Accept: application/json, text/plain, */*" -H "Content-Type: application/json" -H "Origin: https://skills.developer.huaweicloud.com" -H "Referer: https://skills.developer.huaweicloud.com/" -d "{\"skill_id\":\"skills/<category>/<skill-name>\"}"
powershell

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The installation instructions require an unversioned npx skills invocation and automate its confirmation prompts. Although the detected reference is explanatory text, the corresponding executable commands expose the agent to changing installer code without review.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

This installation-option reference leads to execution of the unpinned skills package through npx. It is part of the same supply-chain exposure as the adjacent command, rather than a separate MCP-server vulnerability.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

This command executes the unversioned skills npm package and installs skill content from the mutable master branch. Compromise of the installer distribution or repository could introduce code or agent instructions that run with the agent's privileges.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The ClawHub installation option uses the same unversioned npx installer as the other options. Changing distribution channels does not remove the installer supply-chain risk, and no immutable skill artifact is specified.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The command executes an unpinned npm package to retrieve a skill from a mutable remote URL. A compromised installer release or substituted remote skill could gain execution or influence subsequent agent actions.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The GitHub fallback retains the unversioned npx skills dependency. This repeats the existing supply-chain weakness rather than providing a pinned or independently verified fallback.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

This fallback command executes an unversioned skills package and selects repository content without an immutable revision. An upstream compromise could therefore change both the installer and the installed agent instructions without changes to this file.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The Windows installation option references the same unversioned installer used on Bash. PowerShell argument handling does not mitigate the risk of executing a compromised or unexpectedly changed npm package.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The PowerShell command executes an unversioned npx package and installs from master. These mutable dependencies can introduce malicious code or instructions under the agent's account.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

This Windows ClawHub option repeats the unpinned installer dependency. It is not an MCP server, but the underlying risk of fetching and executing mutable package code is valid.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The command executes the unversioned skills package and fetches skill content without an immutable artifact identifier. Compromise of either dependency can affect the agent's local execution or subsequent behavior.

Content

No source excerpt is available for this finding.