Back to skill

Security audit

全流程会议助理虾

Security checks for vulnerabilities and agentic risk

Overview

This meeting assistant is mostly purpose-aligned, but it includes an unsafe transcription command that could execute code from a crafted audio file path while also handling sensitive meeting and Feishu data.

Review before installing. Only use this with trusted audio file paths, confirm exact Feishu recipients and message contents before sending, and prefer the openclaw-lark tool path over raw curl. The transcription command should be rewritten to pass paths as arguments to a fixed script before using it on untrusted files.

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:37
Finding

Arbitrary Python Code Execution Through Unsafe File-Path Interpolation

Content
View full analysis
' out = '<输出路径>.txt' # 尝试从文件名解析录音开始时间 m = re.search(r'(\d{8})_(\d{6})', wav) ``` ### Technical Analysis The documented workflow directs the agent to replace the audio and output placeholders inside Python source passed to `python3 -c`. The resulting values are therefore interpreted as part of the program rather than supplied as data through command-line arguments or another safe input channel. If an attacker controls the audio filename or path, a single quote in that path can terminate the `wav` string literal. Additional Python statements can then be inserted before the remainder of the generated source is neutralized. Shell metacharacters could create an additional injection boundary depending on how the agent performs placeholder substitution and invokes the command. No escaping, validation, canonicalization, or argument-based invocation is prescribed. Merely quoting the outer shell command with double quotes does not protect the embedded Python single-quoted literals. ### Attack Path 1. An attacker provides or causes the agent to process an audio file whose filename contains a single quote followed by Python statements. 2. The agent substitutes the attacker-controlled path directly into `wav = ''`. 3. The injected single quote terminates the intended Python string. 4. The following filename content becomes executable Python syntax when the generated command is passed to `python3 -c`. 5. The injected code executes with the operating-system privileges and environment of the agent process. 6. The code can read local ...[truncated 1151 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (7)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The trigger list includes broad everyday phrases such as '这段录音' and '开会了', which can cause the skill to activate in situations where the user did not intend transcription, summarization, or downstream task distribution. In this skill, accidental invocation is more dangerous because it can process sensitive meeting audio and potentially lead to extraction and later transmission of action items to third parties.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill description does not prominently warn users that it reads organizational contacts and can send extracted action items to other people via Feishu. That omission undermines informed consent and increases the chance users provide sensitive meeting recordings without realizing the skill will access address-book data and facilitate external disclosure.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The transcription command explicitly sets language='zh', and the surrounding skill description is entirely Chinese-oriented, but there is no opt-in, language selection, or stated limitation that the skill is intended only for Chinese meetings. This can violate language/locale policy by forcing a specific language behavior without user choice.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
89% confidence
Finding

The skill explicitly instructs use of external Feishu API calls, including obtaining a tenant token and interacting with remote services. External transmission is expected for the business purpose, but it still creates a real data-exposure surface because meeting-derived task data and directory lookups leave the local environment and rely on secret-bearing network requests.

Content

Scanner excerpt · SKILL.md (reported line 186)May include surrounding context.

text
  等待用户选择。

**优先级 2:飞书 Message API(curl 直接调用)**

仅当 openclaw-lark 插件不可用且用户选择不安装时使用。

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

This step sends messages to Feishu recipients using open_id, meaning extracted meeting content is transmitted to third parties outside the immediate local workflow. In context, this is more dangerous than generic network use because the skill derives actionable summaries from potentially sensitive recordings, and a recipient-resolution or confirmation mistake can disclose internal plans to unintended users.

Content

Scanner excerpt · SKILL.md (reported line 206)May include surrounding context.

  1. 发送消息:
bash
curl -s -X POST "https://open.feishu.cn/open-apis/im/v1/messages?receive_id_type=open_id" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{

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 · references/feishu-message-api.md (reported line 8)May include surrounding context.

获取 tenant_access_token:

bash
curl -s -X POST https://open.feishu.cn/open-apis/auth/v3/tenant_access_token/internal \
  -H "Content-Type: application/json" \
  -d '{"app_id":"'"$FEISHU_APP_ID"'","app_secret":"'"$FEISHU_APP_SECRET"'"}' \
  | jq -r '.tenant_access_token'

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 · references/feishu-message-api.md (reported line 21)May include surrounding context.

端点: POST /im/v1/messages?receive_id_type=open_id

bash
curl -s -X POST "https://open.feishu.cn/open-apis/im/v1/messages?receive_id_type=open_id" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{