Back to skill

Security audit

Make claw friends and share task with Email, show me your claw business card

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed email-based agent collaboration tool, but it needs review because it can process remote tasks, send mail, mutate a token ledger, and expose mailbox credentials with weak safety controls.

Install only if you are comfortable with a skill that reads and sends mailbox messages, stores email credentials locally, records full task exchanges, and updates a local token ledger. Before use, require owner confirmation for all inbound tasks, outbound tasks, and settlements; use a dedicated low-privilege mailbox; avoid secrets or private data in delegated tasks; and fix credential redaction, message authentication, path validation, and settlement validation.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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 (4)

T01 · Skill Instruction Hijacking

Error
Location
scripts/receive_mail.py:70
Finding

Unauthenticated Email Messages Can Hijack Agent Task Execution

Content
View full analysis
dict: """解析邮件内容,提取 JSON body""" if msg.is_multipart(): for part in msg.walk(): content_type = part.get_content_type() if content_type == "application/json": payload = part.get_payload(decode=True) if payload: return json.loads(payload.decode("utf-8")) # 也尝试 text/plain if content_type == "text/plain": payload = part.get_payload(decode=True) if payload: try: return json.loads(payload.decode("utf-8")) except: pass else: payload = msg.get_payload(decode=True) if payload: try: return json.loads(payload.decode("utf-8")) except: pass return {} ``` `scripts/receive_mail.py:142-159` returns the unverified sender and content as actionable message data: ```python raw_e ...[truncated 2728 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/settle.py:105
Finding

Task ID Path Traversal Enables Filesystem Access Outside the Tasks Directory

Content
View full analysis
dict: """ 从任务记录中获取 Token 消耗 任务记录中应包含 taskStartTokens 和 taskEndTokens """ task_path = workspace / "agent-network" / "tasks" / f"{task_id}.json" if not task_path.exists(): return {"error": f"任务 {task_id} 不存在", "tokenCount": 0} task = json.loads(task_path.read_text(encoding="utf-8")) ``` ### Technical Analysis The `task_id` value originates from command-line input and is interpolated directly into a filesystem path. It is not validated as a UUID and is not checked for path separators or traversal components. `pathlib.Path` does not automatically confine a path beneath its preceding directory. A value containing components such as `../` can therefore resolve outside `agent-network/tasks`. The automatic `.json` suffix limits the target to a corresponding JSON filename but does not prevent traversal. The read-oriented scripts can attempt to parse JSON outside the intended directory. `settle.py` additionally writes the loaded object back after changing task fields, so a traversed target with a compatible structure may be modified. ### A ...[truncated 1212 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/settle.py:45
Finding

Settlement Validation Flaws Permit Ledger Inflation and Duplicate Transactions

Content
View full analysis
0 else amount / (1 + profit_margin) profit_amount = round(base_cost * profit_margin, 6) ``` `scripts/settle.py:86-102` applies the unchecked amount and always appends a new transaction: ```python # 更新余额 if tx_type == "earn": ledger["balance"] = round(ledger["balance"] + amount, 6) ledger["totalEarned"] = round(ledger["totalEarned"] + amount, 6) elif tx_type == "spend": ledger["balance"] = round(ledger["balance"] - amount, 6) ledger["totalSpent"] = round(ledger["totalSpent"] + amount, 6) elif tx_type == "topup": ledger["balance"] = round(ledger["balance"] + amount, 6) ledger["totalEarned"] = round(ledger["totalEarned"] + amount, 6) elif tx_type == "withdraw": ledger["balance"] = round(ledger["balance"] - amount, 6) ledger["totalSpent"] = round(ledger["totalSpent"] + amount, 6) ledger["transactions"].append(tx) ``` ### Technical Analysis Settlement parameters are accepted directl ...[truncated 2076 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/init.py:39
Finding

SMTP and IMAP Credentials Are Stored in Plaintext and Exposed by the Identity Command

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 (29)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The description presents a broad collaboration-network skill with active capabilities: connecting agents via email, managing friends, delegating tasks, receiving tasks, and billing/settlement. The provided code only implements a local inspection utility for existing task records. It reads task JSON files and prints task metadata, logs, results, and billing information, or lists available tasks. While viewing collaboration history and bills is loosely related to a small subset of the declared scenarios, the actual code’s primary purpose is much narrower and does not implement the core network/email/delegation/payment behaviors described. Therefore the description does not accurately represent what this code chunk actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

声明描述的是一个较完整的多 Agent 协作网络技能,核心能力包括通过邮箱连接其他 Agent、建立好友关系、委托任务、协作处理、账单结算和记录管理。实际代码却只是单一的 validate_bill.py 脚本,用于验证一份账单是否在预算范围内、利润率是否合理、以及金额计算是否一致。虽然“查看 Token 账单/结算”与账单相关,代码只能做账单争议校验这一小部分辅助功能,远不足以支撑所声明的主要用途,也没有访问或使用邮箱、联系人、任务分发、结算流水等资源。因此描述与代码行为存在明显且实质性的不匹配。

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill states that sending tasks and confirming bills default to automatic execution, which allows outbound delegation and payment-related actions without explicit user approval. Because the skill uses email as an external communication channel and interacts with billing records, this can cause data exfiltration, unwanted commitments, or fraudulent settlement if triggered maliciously or accidentally.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
85% confidence
Finding

The skill describes capabilities that require file access and network communication, but it does not declare any explicit tool scope or permission boundaries. In an agent framework, missing scope declarations weaken least-privilege controls and make it harder for operators to understand or restrict what the skill may access.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger phrases are broad, natural-language commands like adding friends, outsourcing work, or checking balances, which increases the chance of accidental activation. In this skill, unintended activation is more dangerous because activation can lead to network communication, task delegation, and billing workflows.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The description mentions that all exchanges are recorded, including sent and received mail content, but it does not provide a clear up-front privacy warning near configuration and usage. Since delegated tasks and email contents may include sensitive user data, silent or underemphasized full-content logging increases privacy and retention risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill explicitly describes mutating a local accounting file (ledger.json) and sending billing emails as part of settlement, but does not require an explicit user confirmation or warning before these side effects occur. In an agent-to-agent workflow, that can lead to unauthorized state changes, unintended payments, or outbound communications triggered by another agent or ambiguous user input.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The document explicitly warns not to share SMTP/IMAP configuration and passwords in the public identity card, but then defines a friends.json structure that stores counterparties' SMTP/IMAP host, port, and user details. Even without passwords, centralizing mail infrastructure details and account identifiers for multiple agents increases privacy exposure, enables targeted phishing/account-enumeration, and normalizes collecting more connection metadata than is needed for task routing.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The protocol sends task descriptions, input data, results, billing details, agent IDs, and email addresses over ordinary email JSON messages without any privacy warning, data-minimization guidance, or trust/authentication requirements. In this skill context, agents may forward user-provided content to external agents, so the absence of explicit consent, sensitivity checks, and message protection can lead to unauthorized disclosure of user data, commercially sensitive prompts/results, and billing metadata.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

This markdown file presents the protocol and safety instructions entirely in Chinese, and it does not indicate that the skill is region-specific or that users may choose another language. The policy requires avoiding forced language or locale constraints unless there is explicit opt-in or a documented justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The module docstring presents the script description, usage, and output notes entirely in Chinese, and the script later emits Chinese-only user-facing messages. This imposes a specific language on users without opt-in or documented locale scope, which matches the language/locale policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The error message and billing summary are emitted solely in Chinese, with no user-selectable locale or fallback. For a general-purpose utility, this forces a language choice on users and can violate organizational language/locale requirements.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The module docstring presents the script description and usage entirely in Chinese, which imposes a specific language on users without any opt-in or indication that the tool is intended only for a Chinese-speaking environment. This matches the language/locale policy violation criteria because no alternative language or explicit locale justification is provided.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The script's success and status prints are all emitted in Chinese, so users receive output in a fixed language regardless of preference. Because there is no opt-in, fallback, or stated locale restriction, this is a natural-language policy concern under the language/locale rule.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The instruction tells the user to invoke the next step using a specific Chinese phrase, "说"展示名片"", without offering a language choice or documenting that the skill is region/language-specific. This is a natural-language policy concern because it imposes a locale/language requirement rather than making it optional.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

This code loads IMAP configuration including credentials from agent-network/identity.json, which is a sensitive credential access path. Although the script has a brief docstring about receiving mail, there is no explicit warning, confirmation, or user-facing disclosure that it will read stored email account credentials.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The script fetches full email bodies and then prints subject, sender, message ID, truncated content, and finally the entire EMAILS_JSON payload to stdout. In an agent-network skill, mailbox contents may include delegated tasks, billing details, or other sensitive cross-agent data, so logging this information can leak private data to users, calling processes, logs, or other components that capture stdout.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

This Python file contains natural-language documentation and runtime messages in Chinese only, including the module docstring and error/output text. The policy scope applies to all file types, and there is no indication that the skill is region-specific or that users can opt into another language/locale.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The module docstring and all user-facing output are written in Chinese, which imposes a specific language on users without any opt-in or alternative locale support. Under the policy rule, forcing a language/locale without user choice is a natural-language policy violation unless clearly justified as region-specific.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The script prints task descriptions, result content, billing data, and the full timeline directly to stdout with no redaction, confirmation step, or sensitivity warning. In this skill’s context, those fields can plausibly contain delegated work product, credentials, internal prompts, customer data, or inter-agent communications, so invoking the command can expose sensitive information to anyone with terminal/log access.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

This file contains natural-language instructions and results in Chinese only, including the top-level usage text. Under the policy, forcing a specific language without opt-in is a violation unless the locale restriction is clearly documented and justified, which is not present here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

The natural-language instructions and trigger examples are entirely in Chinese and present the skill behavior as if Chinese is the expected interaction language. There is no opt-in, language choice, or documented locale restriction justifying the language constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The entire skill document is written in Chinese and does not indicate that users may choose another language or locale. Under SQP-3, language constraints should be opt-in or clearly justified as region-specific; neither is stated here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The document presents all instructions and format descriptions in Chinese and does not offer an alternative language or note that the locale is intentionally region-specific. Per the policy rule, forcing a specific language without user opt-in is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

This code file contains natural-language strings that force a specific language/locale for usage instructions and output. Under the policy, language constraints should either be optional for the user or clearly justified as region-specific; neither is indicated here.

Content

No source excerpt is available for this finding.