Back to skill

Security audit

V1.4.17

Security checks for vulnerabilities and agentic risk

Overview

This skill needs review because its automatic document processing can include folders it promises to exclude and potentially share their contents.

Keep automatic delivery disabled until resolved paths enforce the approved-folder boundary. Review the document folders and Feishu recipient, reconcile dependency instructions with actual checks, and back up summaries before bulk repair. Verify the wrappers' hard-coded /Users/home paths. Remove scheduled jobs with openclaw cron remove <id>. All bundled files were inspected; installed dependencies, live OpenClaw permission enforcement and delivery behavior remain unverified. The evidence supports Review, without establishing malicious intent.

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
  • Rogue AgentSelf-Modification, Session Persistence
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • 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 (28)

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 566)May include surrounding context.

md
> **同时**在 `setup.sh` 的建目录段补对应 `mkdir -p`(两处必须同步,否则新装环境扫描范围与声明不符)。

Code capability 'shell' detected in scripts/setup.sh but not covered by declared permissions.

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'shell' capability that is not covered by the declared 'allowed-tools' preapproval. Review the host runtime's actual enforcement policy; runtime permissions are unknown unless separately verified, and this declaration alone does not establish a permission ceiling.

Content

No source excerpt is available for this finding.

The manual installation command bypasses the documented mandatory Python dependency version pins.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The manual installation path installs unpinned Python packages, bypassing the documented version controls. This permits unreviewed dependency changes to execute during installation and process private documents afterward, although no compromised dependency is demonstrated.

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 · SKILL.md (reported line 177)May include surrounding context.

Step 4: 创建目录结构

bash
mkdir -p ~/.openclaw/workspace/knowledge/.analysis/summaries/archives
mkdir -p ~/.openclaw/workspace/knowledge/temp_docs
touch ~/.openclaw/workspace/knowledge/文章目录/文章目录.md

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 · SKILL.md (reported line 178)May include surrounding context.

Step 4: 创建目录结构

bash
mkdir -p ~/.openclaw/workspace/knowledge/.analysis/summaries/archives
mkdir -p ~/.openclaw/workspace/knowledge/temp_docs
touch ~/.openclaw/workspace/knowledge/文章目录/文章目录.md

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/setup.sh (reported line 92)May include surrounding context.

sh
# [1/8] 创建目录
echo ""
echo "[1/8] 创建知识库目录..."
mkdir -p ~/.openclaw/workspace/knowledge/.analysis/summaries/archives
mkdir -p ~/.openclaw/workspace/knowledge/temp_docs
# v1.4.13:部署 cron wrapper(23:00 / 06:00 改 command 型后由 cron 直接调用)
mkdir -p ~/.openclaw/workspace/.analysis/cron_wrappers

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/setup.sh (reported line 93)May include surrounding context.

sh
echo ""
echo "[1/8] 创建知识库目录..."
mkdir -p ~/.openclaw/workspace/knowledge/.analysis/summaries/archives
mkdir -p ~/.openclaw/workspace/knowledge/temp_docs
# v1.4.13:部署 cron wrapper(23:00 / 06:00 改 command 型后由 cron 直接调用)
mkdir -p ~/.openclaw/workspace/.analysis/cron_wrappers
# v1.4.15:建全 knowledge_policy.SCAN_TOP_DIRS 白名单里的 8 个目录。

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/setup.sh (reported line 99)May include surrounding context.

sh
# v1.4.15:建全 knowledge_policy.SCAN_TOP_DIRS 白名单里的 8 个目录。
#   此前只建 4 个 → 新装环境的扫描范围与白名单声明不符
#   (scanner 判定:"folder exclusions ... weaker than documented")
mkdir -p ~/.openclaw/workspace/knowledge/"Macro Financials"
mkdir -p ~/.openclaw/workspace/knowledge/工程效能专家
mkdir -p ~/.openclaw/workspace/knowledge/房产效能专家
mkdir -p ~/.openclaw/workspace/knowledge/合同对比

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/setup.sh (reported line 100)May include surrounding context.

sh
#   此前只建 4 个 → 新装环境的扫描范围与白名单声明不符
#   (scanner 判定:"folder exclusions ... weaker than documented")
mkdir -p ~/.openclaw/workspace/knowledge/"Macro Financials"
mkdir -p ~/.openclaw/workspace/knowledge/工程效能专家
mkdir -p ~/.openclaw/workspace/knowledge/房产效能专家
mkdir -p ~/.openclaw/workspace/knowledge/合同对比
mkdir -p ~/.openclaw/workspace/knowledge/文章目录

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/setup.sh (reported line 101)May include surrounding context.

sh
#   (scanner 判定:"folder exclusions ... weaker than documented")
mkdir -p ~/.openclaw/workspace/knowledge/"Macro Financials"
mkdir -p ~/.openclaw/workspace/knowledge/工程效能专家
mkdir -p ~/.openclaw/workspace/knowledge/房产效能专家
mkdir -p ~/.openclaw/workspace/knowledge/合同对比
mkdir -p ~/.openclaw/workspace/knowledge/文章目录
mkdir -p ~/.openclaw/workspace/knowledge/知识库搭建手册

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/setup.sh (reported line 102)May include surrounding context.

sh
mkdir -p ~/.openclaw/workspace/knowledge/"Macro Financials"
mkdir -p ~/.openclaw/workspace/knowledge/工程效能专家
mkdir -p ~/.openclaw/workspace/knowledge/房产效能专家
mkdir -p ~/.openclaw/workspace/knowledge/合同对比
mkdir -p ~/.openclaw/workspace/knowledge/文章目录
mkdir -p ~/.openclaw/workspace/knowledge/知识库搭建手册
mkdir -p ~/.openclaw/workspace/knowledge/临时

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/setup.sh (reported line 103)May include surrounding context.

sh
mkdir -p ~/.openclaw/workspace/knowledge/工程效能专家
mkdir -p ~/.openclaw/workspace/knowledge/房产效能专家
mkdir -p ~/.openclaw/workspace/knowledge/合同对比
mkdir -p ~/.openclaw/workspace/knowledge/文章目录
mkdir -p ~/.openclaw/workspace/knowledge/知识库搭建手册
mkdir -p ~/.openclaw/workspace/knowledge/临时
touch ~/.openclaw/workspace/knowledge/文章目录/文章目录.md

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/setup.sh (reported line 104)May include surrounding context.

sh
mkdir -p ~/.openclaw/workspace/knowledge/房产效能专家
mkdir -p ~/.openclaw/workspace/knowledge/合同对比
mkdir -p ~/.openclaw/workspace/knowledge/文章目录
mkdir -p ~/.openclaw/workspace/knowledge/知识库搭建手册
mkdir -p ~/.openclaw/workspace/knowledge/临时
touch ~/.openclaw/workspace/knowledge/文章目录/文章目录.md
echo "✅ 目录创建完成"

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/setup.sh (reported line 105)May include surrounding context.

sh
mkdir -p ~/.openclaw/workspace/knowledge/合同对比
mkdir -p ~/.openclaw/workspace/knowledge/文章目录
mkdir -p ~/.openclaw/workspace/knowledge/知识库搭建手册
mkdir -p ~/.openclaw/workspace/knowledge/临时
touch ~/.openclaw/workspace/knowledge/文章目录/文章目录.md
echo "✅ 目录创建完成"

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/setup.sh (reported line 182)May include surrounding context.

sh
done

# 记录本机实际依赖版本 → 供升级时核对供应链
mkdir -p ~/.openclaw/workspace/.analysis
{
  echo "# macmini-knowledge-base v1.4.17 实测通过的依赖版本"
  echo "# 生成时间:$(date '+%Y-%m-%d %H:%M:%S')"

A crashed analysis process can be reported as successful despite the explicit promise that crashes return failure.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The wrapper captures the analysis exit status but does not enforce it when a statistics line exists. A process that emits statistics and subsequently fails can therefore produce a successful cron result, undermining monitoring integrity and concealing incomplete processing.

Content

No source excerpt is available for this finding.

The claimed directory-whitelist boundary can be bypassed by symlinks to non-whitelisted directories inside knowledge/.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The realpath check confines resolved files to knowledge/ but does not enforce the approved top-level directories or exclusions on resolved targets. A symlink inside an approved directory can expose documents from an unapproved internal directory to analysis and scheduled Feishu delivery, violating the declared confidentiality boundary.

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/setup.sh (reported line 95)May include surrounding context.

sh
mkdir -p ~/.openclaw/workspace/knowledge/.analysis/summaries/archives
mkdir -p ~/.openclaw/workspace/knowledge/temp_docs
# v1.4.13:部署 cron wrapper(23:00 / 06:00 改 command 型后由 cron 直接调用)
mkdir -p ~/.openclaw/workspace/.analysis/cron_wrappers
# v1.4.15:建全 knowledge_policy.SCAN_TOP_DIRS 白名单里的 8 个目录。
#   此前只建 4 个 → 新装环境的扫描范围与白名单声明不符
#   (scanner 判定:"folder exclusions ... weaker than documented")

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/setup.sh (reported line 314)May include surrounding context.

sh
if [[ $REPLY =~ ^[Yy]$ ]]; then
    read -p "  请输入飞书 webhook URL: " FEISHU_WEBHOOK_URL
    if [[ -n "$FEISHU_WEBHOOK_URL" ]]; then
        mkdir -p ~/.openclaw/workspace/.feishu
        # webhook 是 bearer 型凭证:以受限 umask 写入,并显式 chmod 600
        ( umask 077; echo "$FEISHU_WEBHOOK_URL" > ~/.openclaw/workspace/.feishu/webhook_url )
        chmod 600 ~/.openclaw/workspace/.feishu/webhook_url

Homebrew installation does not perform the post-install version validation promised by the documentation.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The installer compares versions only for already-installed packages, and neither installation nor upgrade is followed by validation against the tested version list. Because Homebrew installs its currently available release, setup can accept untested document-processing dependencies while implying that version validation occurred; recording versions afterward does not enforce the security boundary.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/utils.py (reported line 346)May include surrounding context.

python
def extract_doc_via_antiword(filepath):
    try:
        result = subprocess.run(['antiword', filepath], capture_output=True, timeout=10)
        if result.returncode == 0:
            text = result.stdout.decode('utf-8', errors='replace')
            if len(text) > 100:

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/utils.py (reported line 359)May include surrounding context.

python
def _kill_proc_tree(pid):
    try:
        # v1.4.3: 替换 os.system 为 subprocess.run(避免 shell 注入 + 抑制输出)
        subprocess.run(["pkill", "-P", str(pid)],
                       capture_output=True, timeout=5)
    except subprocess.TimeoutExpired:
        pass  # pkill 超时属预期,继续走下面的 kill

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/utils.py (reported line 388)May include surrounding context.

python
except OSError as e:
                # 残留的旧转换产物删不掉:记录,避免静默堆积
                print(f"[临时文件清理失败] {tmp_out}: {e}", file=sys.stderr)
        proc = subprocess.Popen(
            ['soffice', '--headless', '--convert-to', new_ext, '--outdir', tmp_dir, filepath],
            stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL
        )

Several extraction paths bypass clean_noise despite the manifest's explicit claim of complete coverage.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The manifest states that all extraction paths, including DOCX, XLSX, PPTX, and OCR, pass through clean_noise. However, the native fallback returns at L501, L517, and L535 return uncleaned text; the Office OCR return at L645 and legacy XLS return at L682 also bypass cleaning. Consequently, the promised removal of disclaimer noise is not consistently applied to documents entering the knowledge base.

Content

No source excerpt is available for this finding.

PPTX fallback silently limits extraction to the first 20 slides despite the full-document extraction claim.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The manifest advertises complete extraction of long documents using a 500,000-character limit instead of short hard truncations. The native PPTX fallback instead selects only prs.slides[:20], excluding later slides regardless of the extracted character count. This can leave substantial document content absent from subsequent summaries and RAG indexing without indicating that extraction was partial.

Content

No source excerpt is available for this finding.