Back to skill

Security audit

code-review

Security checks for vulnerabilities and agentic risk

Overview

This skill's review workflows can delete local work, execute repository-controlled code, and publish changes without clear user control.

Review before installing. Use disposable, uniquely identified checkouts inside a sandbox without personal credentials, and prefer repository-scoped read-only access. Require explicit authorization for destructive cleanup, remote publication, configuration changes, and unrelated skill updates. Keep review tools separate from development environments and verify downloaded installers. The scanner's partial status comes from an absent runtime-generated ignored_issues.md; the FAQ and bundled template explain its creation, resolving that gap. External installers and future repository code were not executed or independently verified.

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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
references/language_checks/typescript_javascript_check.md
Finding

Untrusted repositories supply executable review commands without a host-security boundary

Content
View full analysis

Impact

An attacker-controlled repository can run arbitrary commands with the review process's permissions, including reading accessible credentials, changing files outside the checkout, or making network requests.

T09 · Insecure Skill Coding Practices

Warning
Location
references/repository_access.md
Finding

Repository-name collisions can direct destructive updates and report pushes at the wrong repository

Content
View full analysis

Impact

The Skill can discard another checkout's uncommitted changes and untracked files, review the wrong source, and push generated reports to the wrong remote using the Agent's credentials.

Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • 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
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
Findings (52)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Version management for code-review is consistent with its declared maintenance scope. However, openclaw skill update-all affects all unpinned skills in the session rather than only code-review. This broader skill-management capability is not declared and is not necessary merely to perform code reviews or maintain this individual skill. The chunk presents command templates; it does not demonstrate that they have been executed.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

Updating @tinycen/code-review matches the declared self-update capability. However, openclaw skills update --all broadens the operation to every installed workspace skill. Managing unrelated skills is not declared and exceeds a supporting implementation detail of code review or self-update.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

Ignoring node_modules can support code-review preparation, but committing and pushing repository changes goes beyond local review setup. The declared purpose covers repository cloning, analysis, and report distribution, not modifying and publishing the reviewed repository's history.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

忽略 .venv 可以是依赖安装的辅助步骤,但自动提交并推送 .gitignore 变更超出了所声明的代码检查与报告处理范围。报告分发并不等同于向源代码仓库发布配置变更;git push 还可能推送其他尚未发布的提交。

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

Pulling repository updates is a reasonable supporting operation for code review. However, git reset --hard HEAD and git clean -fd also destroy local work. This destructive behavior is not disclosed by the declared review and reporting purpose, and can remove changes that would otherwise be reviewed.

Content

No source excerpt is available for this finding.

External Script Fetching

High
Category
Supply Chain
Confidence
98% confidence
Finding

The command immediately executes a mutable remote installer with the agent's privileges, without reviewing or verifying its contents. The official HTTPS endpoint reduces some risks, but compromise of the distribution service could still provide arbitrary code; the remote workflow explicitly automates this installation.

Content

Scanner excerpt · references/frontend_dependency_installation/node_environment.md (reported line 56)May include surrounding context.

  • Unix/Linux/macOS:
    bash
    curl https://get.volta.sh | bash
    
  • Windows / PowerShell:
    powershell

Chaining Abuse

High
Category
Tool Misuse
Confidence
98% confidence
Finding

Piping the downloaded response directly into bash crosses the network-to-execution trust boundary without an inspection or verification step. This is the same underlying vulnerability as the external-script-fetching finding, not evidence of a separate malicious chaining technique.

Content

Scanner excerpt · references/frontend_dependency_installation/node_environment.md (reported line 56)May include surrounding context.

  • Unix/Linux/macOS:
    bash
    curl https://get.volta.sh | bash
    
  • Windows / PowerShell:
    powershell

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · references/installation.md (reported line 163)May include surrounding context.

mkdir -p download_skills curl -L -o download_skills/code-review.zip "https://clawhub.ai/api/skills/download?slug=code-review&ownerHandle=tinycen" unzip download_skills/code-review.zip -d download_skills/code-review rm download_skills/code-review.zip

text

2. **移动到对应技能目录**:

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · references/installation.md (reported line 235)May include surrounding context.

mkdir -p download_skills curl -L -o download_skills/code-review.zip "https://modelscope.cn/skills/tinycen/code-review/archive/zip/master.zip" unzip download_skills/code-review.zip -d download_skills/code-review rm download_skills/code-review.zip

text

解压后,将包含 `SKILL.md` 的技能目录移动到你所用工具的技能目录中,具体步骤与「场景二 > 首次安装」步骤 2 相同。更新时重新下载并覆盖旧目录即可。

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/repository_access.md (reported line 14)May include surrounding context.

md
首次从 GitHub 拉取/推送代码前,必须执行以下检查流程,确保 SSH 认证可用:

1. **检查本地是否存在 SSH 密钥**
   - 检查默认密钥路径:`~/.ssh/id_ed25519.pub` 或 `~/.ssh/id_rsa.pub`
   - 若存在:进入步骤 2
   - 若不存在:进入步骤 3(自动生成)

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/repository_access.md (reported line 28)May include surrounding context.

bash
     ssh-keygen -t ed25519 -C "code-review@automation" -f ~/.ssh/id_ed25519 -N ""
     ```
   - 读取公钥内容:`cat ~/.ssh/id_ed25519.pub`
   - **将公钥内容发送给用户**,并提示:
     > 检测到本地未配置 GitHub SSH 密钥,已自动生成。请将以下公钥添加到 GitHub 账号的 **Settings → SSH and GPG keys → New SSH key** 中,添加完成后告知我,我将继续执行。\n\n`<公钥内容>`
   - 等待用户确认已添加后,重新执行步骤 2

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/repository_access.md (reported line 36)May include surrounding context.

md
4. **已有密钥但权限不足(GitHub SSH 权限重试)**
   - 当克隆或推送因 `Permission denied (publickey)` 或 `Could not read from remote repository` 失败时:
     - **不重新生成 SSH 密钥**(避免之前的配置失效)
     - 读取当前使用的公钥内容(`~/.ssh/id_ed25519.pub` 或 `~/.ssh/id_rsa.pub`)
     - **将公钥内容发送给用户**,并提示:
       > 当前 SSH 密钥无法访问目标仓库,可能是该仓库需要额外权限,或密钥尚未添加到 GitHub 账号。请将以下公钥添加到 GitHub 账号的 **Settings → SSH and GPG keys → New SSH key** 中,或确认该密钥对目标仓库有访问权限。处理完成后告知我,我将继续执行。\n\n`<公钥内容>`
     - 等待用户确认后,重新尝试失败的 Git 操作

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/repository_access.md (reported line 14)May include surrounding context.

md
首次从 GitHub 拉取/推送代码前,必须执行以下检查流程,确保 SSH 认证可用:

1. **检查本地是否存在 SSH 密钥**
   - 检查默认密钥路径:`~/.ssh/id_ed25519.pub` 或 `~/.ssh/id_rsa.pub`
   - 若存在:进入步骤 2
   - 若不存在:进入步骤 3(自动生成)

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/repository_access.md (reported line 36)May include surrounding context.

md
4. **已有密钥但权限不足(GitHub SSH 权限重试)**
   - 当克隆或推送因 `Permission denied (publickey)` 或 `Could not read from remote repository` 失败时:
     - **不重新生成 SSH 密钥**(避免之前的配置失效)
     - 读取当前使用的公钥内容(`~/.ssh/id_ed25519.pub` 或 `~/.ssh/id_rsa.pub`)
     - **将公钥内容发送给用户**,并提示:
       > 当前 SSH 密钥无法访问目标仓库,可能是该仓库需要额外权限,或密钥尚未添加到 GitHub 账号。请将以下公钥添加到 GitHub 账号的 **Settings → SSH and GPG keys → New SSH key** 中,或确认该密钥对目标仓库有访问权限。处理完成后告知我,我将继续执行。\n\n`<公钥内容>`
     - 等待用户确认后,重新尝试失败的 Git 操作

Credential Access

High
Category
Privilege Escalation
Confidence
90% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/repository_access.md (reported line 26)May include surrounding context.

md
3. **自动生成 SSH 密钥(无密钥时)**
   - 生成 ed25519 密钥对:
     ```bash
     ssh-keygen -t ed25519 -C "code-review@automation" -f ~/.ssh/id_ed25519 -N ""
     ```
   - 读取公钥内容:`cat ~/.ssh/id_ed25519.pub`
   - **将公钥内容发送给用户**,并提示:

Repository updates unconditionally discard local changes and delete untracked files.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

The update procedure mandates discarding tracked modifications and deleting untracked files without checking for valuable local work or obtaining approval. A code review does not require these destructive operations, and deleted untracked content may be unrecoverable.

Content

No source excerpt is available for this finding.

The update workflow discards local changes and deletes untracked files without warning the user about data loss.

High
Category
Output Handling
Confidence
99% confidence
Finding

The instructions present destructive cleanup as a routine prerequisite without warning that it removes user work. Users requesting a review could therefore lose local modifications or untracked files without knowingly authorizing their deletion.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
99% confidence
Finding

The mandated git reset --hard HEAD overwrites tracked working-tree changes during an otherwise observational review task. Combined with the following cleanup command, it can destroy user work without consent.

Content

Scanner excerpt · references/repository_access.md (reported line 84)May include surrounding context.

md
当本地已存在目标仓库时:

1. 进入该仓库目录
2. 执行 `git reset --hard HEAD` 重置本地修改
3. 执行 `git clean -fd` 清理未跟踪文件
4. 执行 `git pull origin main` 拉取最新代码

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
99% confidence
Finding

The command block performs destructive reset and cleanup with no dirty-tree guard or confirmation. Its cd command is also not explicitly failure-guarded, so a shell continuing after a failed directory change could apply those operations to a different current repository.

Content

Scanner excerpt · references/repository_access.md (reported line 92)May include surrounding context.

bash
cd <workspace>/repos/<域名>/<仓库名>
git reset --hard HEAD
git clean -fd
git pull origin main

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/review_process.md (reported line 281)May include surrounding context.

md
- **检查项**:元数据是否完整

#### 环境文件
- **检查项**:.env 文件是否被 .gitignore 排除
- **检查项**:环境变量模板文件(.env.example)是否提供

#### 其他配置文件

Scheduled reviews discard uncommitted repository changes, an unjustified destructive capability for a code-review skill.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

The scheduled workflow explicitly discards uncommitted changes before checking versions. This can destroy work in either repository even when no new version exists and no review will subsequently run; unattended execution increases the risk.

Content

No source excerpt is available for this finding.

Scheduled reviews discard uncommitted changes without warning about loss of user work.

High
Category
Output Handling
Confidence
99% confidence
Finding

The workflow provides no data-loss warning or approval gate before reverting uncommitted changes. Scheduling a review does not itself authorize deleting later local work, especially across two repositories.

Content

No source excerpt is available for this finding.

Scheduled code review automatically discards local work, an unjustified destructive capability for a review tool.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

The scheduled workflow explicitly resets tracked files and deletes untracked files and directories whenever local changes exist, without confirming that the checkout is disposable. An unattended code-review task can therefore destroy user work unrelated to the review, even before determining whether a new review is needed.

Content

No source excerpt is available for this finding.

The scheduled workflow automatically discards local changes and deletes untracked files without a data-loss warning.

High
Category
Output Handling
Confidence
99% confidence
Finding

The workflow mandates automatic deletion of local changes without a data-loss warning or approval gate. Scheduling a review is not informed consent to discard tracked edits and untracked files, and unattended execution increases the likelihood that deletion occurs unnoticed.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
99% confidence
Finding

The instruction explicitly invokes git reset --hard HEAD, which discards staged and unstaged tracked-file changes, together with git clean -fd, which removes untracked files and directories. These destructive operations exceed what is necessary for code review and can cause unrecoverable loss in a reused working directory.

Content

Scanner excerpt · workflows/scheduled_workflow.md (reported line 28)May include surrounding context.

md
- 进入本地绝对路径。
- 检查本地仓库是否有未提交的修改(`git status --short`)。
  - 若有修改:自动执行 `git reset --hard HEAD` 与 `git clean -fd` 撤销所有本地修改。
- 检查仓库是否存在 tag 标签:
  - 存在 tag:获取本地最新 tag,进入步骤 3。
  - 不存在 tag:获取本地 HEAD,进入步骤 4。