T07 · Tool Hijacking and Spoofing
- Location
scripts/brief.py:62- Finding
Execution of Unverified External Helper Scripts
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
The skill mostly matches its sales-brief purpose, but it can modify business pipeline data and runs unverified helper scripts outside the reviewed package.
Review before installing if this would hold real customer, contract, or revenue data. Use it only with non-sensitive or backed-up pipeline data, and be aware that write and search operations depend on other local scripts outside this reviewed package. The evidence supports Review rather than a malicious classification because the risky behavior is largely tied to the stated sales workflow and no intentional exfiltration, deception, or destructive payload was found.
scripts/brief.py:62Execution of Unverified External Helper Scripts
scripts/brief.py:38Pipeline Corruption Can Cause Silent Replacement with Sample Data
The skill documents execution of a local Python script and use of shell/file capabilities, but it does not declare any explicit tool scope such as allowed-tools or permissions. This creates an authorization ambiguity where an agent may invoke broader local capabilities than users expect, increasing the risk of unintended file access or command execution.
The trigger phrases include very common terms like '销售简报', '每日报告', 'pipeline', 'daily brief', and 'sales report', which are likely to match ordinary user conversations unrelated to this specific skill. Overbroad activation can cause the skill to run unexpectedly, exposing or modifying locally stored customer, contract, and pipeline data without clear user intent.
The skill description explains how to add clients, contracts, and leads to a local JSON database, but it does not clearly warn users that potentially sensitive customer and contract information will be stored on disk under the user's home directory. In a sales context, this data may include confidential commercial information, so silent local persistence increases privacy, compliance, and accidental disclosure risks.
The file's natural-language interface, usage text, and generated output are entirely in Chinese, which effectively imposes a language choice on users. There is no indication that the skill supports language selection or that the Chinese-only constraint is explicitly documented as a justified regional requirement.
The script persists customer, contract, lead, and revenue data to disk under the user's home directory without any prominent notice, consent flow, retention policy, or protection controls. Because this is commercially sensitive data, undisclosed local persistence can cause confidentiality loss through local compromise, backups, shared accounts, or later access by other tools.
The skill exceeds its apparent scope by invoking external helper scripts, including one for file writes and another for search, which expands execution capability and trust to code outside this file. In a skill handling customer and contract data, hidden delegation to other scripts increases the attack surface and can lead to unexpected code execution or data exposure if those helpers are modified or untrusted.
The skill performs file writes by executing a separate script without clearly informing the user that another executable component will run. This hidden helper execution is dangerous because it broadens the trust boundary: if the helper is altered, replaced, or behaves unexpectedly, sensitive sales data could be modified, copied, or written elsewhere.
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
try:
write_script = Path.home() / "Library/Application Support/QClaw/openclaw/config/skills/qclaw-text-file/scripts/write_file.py"
result = subprocess.run(
["/opt/homebrew/bin/python3.11", str(write_script),
"--path", str(PIPELINE_FILE),
"--content-file", tmp_path],
The competitor-news feature can trigger external searches via another script, but the user is not clearly warned that network requests may be made or what query terms will be sent. In a sales workflow, even outbound searches can leak strategic interest patterns or internal competitor lists to external services, creating privacy and business-intelligence exposure.
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
for competitor in competitors[:3]: # 只搜索前3个竞品,避免耗时过长
query = f"{competitor} GPU cloud AI算力 新闻"
try:
result = subprocess.run(
["/opt/homebrew/bin/python3.11", str(pansxng_script), query, "--days", str(days)],
capture_output=True,
text=True,
This JSON file stores user-facing text fields such as company names, stages, notes, sources, and priorities entirely in Chinese, with no indication that the skill is region-specific or that users can opt into this locale. Under the stated policy, forcing a specific language without documented opt-in or justification can be a natural-language policy violation.
The manifest emphasizes automatic summarization of follow-up customers, expiring contracts, leads, and competitor dynamics, plus maintaining pipeline data at a high level. The code exposes broad record-management operations including add, update, and full listing of clients, contracts, and leads, making it a more general pipeline database utility than just a brief generator.