Back to skill

Security audit

search-hotel

Security checks for vulnerabilities and agentic risk

Overview

The skill has a coherent hotel-search purpose, but it tells agents to repeatedly run unpinned latest third-party CLI packages while handling an API key.

Install only if you are comfortable with the RollingGo CLI package being fetched or upgraded to the newest release at runtime. Prefer a pinned reviewed version, avoid global upgrades, inject only the RollingGo key for this skill, avoid command-line API-key arguments, and do not enable broad shell environment import unless you understand the credential exposure.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (2)

T08 · Insecure Dependencies

Error
Location
SKILL.md:58
Finding

Unpinned Third-Party Packages Are Downloaded and Executed Automatically

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:58-68; references/rollinggo-npx.md:21-33,49-57; references/rollinggo-uv.md:23-40,50-58
Vulnerability Type: Unpinned dependency execution and supply-chain exposure
Risk Level: High

Vulnerable Code Snippets

SKILL.md:58-68:

markdown
## 版本新鲜度(始终使用最新版)

本技能默认策略:每次执行都使用最新发布版本。

- **npm/npx:** `npx --yes --package rollinggo@latest rollinggo ...`
- **uvx:** `uvx --refresh --from rollinggo@latest rollinggo ...`

如果使用已安装命令而不是临时执行,先升级再运行:

- **npm 全局:** `npm install -g rollinggo@latest`
- **uv 工具:** `uv tool upgrade rollinggo@latest`

references/rollinggo-npx.md:21-33:

markdown
### 临时执行(npx — 无需安装)

```bash
npx --yes --package rollinggo@latest rollinggo --help
npx --yes --package rollinggo@latest rollinggo search-hotels --origin-query "..." --place "东京迪士尼" --place-type "<查看 --help 获取合法值>"

全局安装(适合长期高频使用)

bash
npm install -g rollinggo@latest
rollinggo --help
text

`references/rollinggo-uv.md:23-40`:

```markdown
### 临时执行(uvx — 无需安装)

> 注意:包名和命令名都叫 `rollinggo`,所以需要 `--from` 语法。
> 显式使用 `rollinggo@latest`,让 uv 解析当前最新发布版本。

```bash
uvx --refresh --from rollinggo@latest rollinggo --help
uvx --refresh --from rollinggo@latest rollinggo search-hotels \
  --origin-query "查找东京迪士尼附近的酒店" \
  --place "东京迪士尼" --place-type "<查看 --help 获取合法值>"

安装为工具(适合长期高频使用)

bash
uv tool install rollinggo@latest
uv tool upgrade rollinggo@latest
rollinggo --help
text

### Technical Analysis

The Skill explicitly requires every execution to resolve or upgrade to `rollinggo@latest`. No exact package version, lockfile, package digest, integrity hash, trusted mirror, or update-review mechanism is provided. Consequently, the code executed at runtime can differ from the code that existed when the Skill was audited.

The `npx --yes` option suppresses the normal installation confirmat
...[truncated 2044 chars]
Remediation
View remediation

Remediation Suggestions

  1. Replace every rollinggo@latest reference with an exact, reviewed version, such as rollinggo@1.0.0.
  2. Pin and verify package integrity using npm lockfiles and integrity metadata or the corresponding Python lockfile and package hashes.
  3. Remove --yes from npx commands so unexpected installation is not silently approved.
  4. Remove --refresh and mandatory upgrade instructions from routine Skill execution.
  5. Separate dependency updates from runtime operation. Review, test, and approve each new package release before changing the pinned version.
  6. Prefer a trusted internal registry or artifact repository containing reviewed package artifacts.
  7. Execute the CLI in a restricted sandbox with only the required API key, limited filesystem access, and constrained network access.
  8. Avoid global installation where possible because it modifies shared executable state.

T09 · Insecure Skill Coding Practices

Warning
Location
references/claw-host-env.md:23
Finding

Fallback Configuration Exposes API Keys Through Shared Host and Shell Environments

Content
View full analysis

Vulnerability Details

File Location: references/claw-host-env.md:23-41
Vulnerability Type: Overbroad secret injection and plaintext credential storage
Risk Level: Medium

Vulnerable Code Snippet

references/claw-host-env.md:23-41:

markdown
### 宿主级注入(多个 skill 共用同一个 key 时)

```json
{ "env": { "RollingGo_API_KEY": "YOUR_KEY" } }

或写入宿主的 .env 文件:~/.openclaw/.env(macOS/Linux)/ %USERPROFILE%\.openclaw\.env(Windows)。
路径覆盖:OPENCLAW_HOME、OPENCLAW_STATE_DIR 或 OPENCLAW_CONFIG_PATH。

Shell 导入兜底

json
{ "env": { "shellEnv": { "enabled": true, "timeoutMs": 15000 } } }

仅当 key 存在于登录 shell 但子进程拿不到时使用。

Sandbox 提醒

text

### Technical Analysis

The Skill legitimately needs access to `RollingGo_API_KEY`, and the same document recommends per-Skill injection as the preferred option. However, the fallback instructions allow the key to be injected at the host level, persisted in a plaintext `.env` file, or imported from the login-shell environment.

Host-level injection can make the RollingGo credential available to unrelated Skills or child processes. Shell environment import can expand exposure further by importing unrelated credentials and sensitive variables from the user's login environment. Plaintext `.env` storage also creates a persistent credential artifact whose protection depends on filesystem permissions, backup handling, and host configuration.

These alternatives exceed the minimum privileges needed for hotel searches. The minimum necessary design is to provide only `RollingGo_API_KEY` to only the `rollinggo-searchhotel` Skill for the duration of its execution.

The repository does not contain code that reads arbitrary credential files or exfiltrates existing secrets. The risk arises from the documented configuration choices and from any process that can subsequently inspect the shared environment or plaintext file.

### Attack Path

1. A user follows the fallbac
...[truncated 1528 chars]
Remediation
View remediation

Remediation Suggestions

  1. Require per-Skill secret injection and remove host-wide environment injection from normal setup guidance.
  2. Disable login-shell environment import for this Skill. Explicitly inject only RollingGo_API_KEY.
  3. Store the key in the host platform's secret manager rather than a plaintext .env file.
  4. If file-based storage is unavoidable, use a dedicated file containing only this key, restrict access to the owning account, and document secure permissions such as mode 0600 on supported systems.
  5. Do not pass API keys through --api-key command-line arguments because process listings, shell history, or logs may expose them.
  6. Ensure sandbox configuration passes only the required key and does not inherit the complete host environment.
  7. Rotate the key after suspected disclosure and support scoped, revocable, and rate-limited credentials where the service permits.
  8. Document that secret values must not be committed to source control, included in diagnostic output, or copied into shared configuration.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (21)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The manifest description and the skill content are written entirely in Chinese and describe the skill behavior without indicating that other languages are supported or that the user can choose a preferred language. This can violate a language/locale policy when a skill implicitly constrains interaction to a single language without opt-in or justification.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The metadata declares primaryEnv/required env as RollingGo_API_KEY, while the operational instructions repeatedly require ROLLINGGO_API_KEY. This inconsistency can cause the skill to miss the configured credential, fail open/closed unpredictably, or encourage users to duplicate secrets under multiple variable names, increasing secret-handling mistakes and operational confusion.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The skill instructs use of rollinggo@latest, which is explicitly unpinned and causes every execution to trust whatever version is current at runtime. This creates a supply-chain risk: a compromised upstream package, malicious release, or breaking change could execute arbitrary code in the agent environment while handling API keys and user queries.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The uvx --refresh --from rollinggo@latest instruction pulls and executes the newest package version on each run, eliminating reproducibility and exposing the environment to upstream package compromise. In this skill context, that risk is amplified because the tool is expected to access an API key and process user-supplied search parameters.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The documentation instructs users to run npx --yes --package rollinggo@latest, which fetches and executes whatever package version is latest at runtime. This creates a supply-chain risk: if the package is compromised, typo-squatted, or a malicious version is published, the skill will execute unreviewed code immediately.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

This example again relies on npx with rollinggo@latest, causing dynamic download and execution of the newest package version. In an agent skill context, that is especially risky because repeated executions can silently change behavior or introduce malicious code without any skill update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The 'default latest-version prefix' explicitly standardizes npx --yes --package rollinggo@latest rollinggo, normalizing execution of unpinned remote code. This increases the chance that all users of the skill will adopt an unsafe update channel, magnifying supply-chain exposure.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The documentation shows --api-key YOUR_API_KEY on the command line without warning that command-line arguments can be exposed through shell history, process listings, logs, and telemetry. In this skill context, the key grants access to a live external service, so accidental disclosure can lead to unauthorized API usage and possible account abuse.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

This reference continues recommending the @latest execution model, which means the command behavior and code trust boundary are uncontrolled. In a tool that may handle API keys and external requests, compromise of the package can lead to credential theft, command abuse, or data exfiltration.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The troubleshooting guidance directs users back to npx --yes --package rollinggo@latest, reinforcing a pattern of executing mutable third-party code during error recovery. Attackers often target such trusted recovery paths because users are more likely to run them without scrutiny.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The troubleshooting note advises injecting the same API key into a host configuration layer but gives no guidance on secure storage, scoping, rotation, or log redaction. That increases the risk of credential persistence in plaintext config files, broader-than-needed exposure across components, and leakage via host diagnostics.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

All instructions, examples, and user-facing query text are presented only in Chinese, and the file does not indicate that Chinese is optional or required for a region-specific reason. Under the stated policy, forcing a specific language without user opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The document instructs users to execute uvx --refresh --from rollinggo@latest rollinggo ..., which fetches and runs the latest published package version at execution time. Using @latest rather than a pinned immutable version creates a supply-chain risk: if the upstream package is compromised or a malicious release is published, users will automatically execute attacker-controlled code.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

This example again directs runtime installation/execution from rollinggo@latest, causing every invocation to trust the current upstream release. In a skill context, this is especially risky because the agent may follow the documented command path repeatedly, expanding exposure to package hijacking or malicious updates.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
99% confidence
Finding

The 'version freshness' section explicitly sets a policy of always using the newest PyPI release. That guidance normalizes execution of unreviewed upstream code and materially increases supply-chain compromise risk if the package account, dependency chain, or publishing pipeline is abused.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The hotel-tags example continues the same unsafe pattern of temporary execution from rollinggo@latest. Even though the command appears read-only, installing and invoking the package still runs arbitrary Python package code during setup/execution if the upstream artifact is malicious.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The troubleshooting section recommends falling back to uvx --refresh --from rollinggo@latest ..., which preserves the unsafe auto-update behavior during failure handling. This is dangerous because users under troubleshooting pressure are more likely to run commands without scrutiny, increasing the chance of executing a compromised release.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The local development section uses uvx --refresh --from . rollinggo --help, which executes directly from the current directory source tree. While common in development, it can still be risky if the repository contents are untrusted or modified, causing local arbitrary code execution under the developer's account.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The documentation shows --api-key YOUR_API_KEY directly in shell command examples without warning that command-line secrets may be exposed via shell history, process listings, logs, or telemetry. This can lead to inadvertent credential leakage, especially when users copy-paste examples verbatim.

Content

No source excerpt is available for this finding.