ClawHub
Back to skill

Security audit

Bankai Skill

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed DeepSeek-backed banking document drafting skill, with the main caution that sensitive banking inputs may be sent to DeepSeek unless a private endpoint is configured.

Install only if you are comfortable using your own DeepSeek API key and sending draft inputs to DeepSeek by default. For customer, account, transaction, regulatory, personnel, or internal bank data, configure a private OpenAI-compatible endpoint with --base-url and manually review all generated drafts before use.

SkillSpector

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 (2)

Vague Triggers

Medium
Confidence
80% confidence
Finding
The skill advertises activation through broad natural-language requests such as '帮我写…' without defining a clear invocation boundary, which can cause the agent to trigger during ordinary conversation or when untrusted content includes similar phrases. In a banking-writing context, accidental activation could route sensitive business details into a document-generation workflow or external model call unexpectedly, increasing confidentiality and workflow-integrity risk.

External Transmission

Medium
Category
Data Exfiltration
Content
const __dirname = dirname(fileURLToPath(import.meta.url));
const { SCENARIOS } = await import(pathToFileURL(join(__dirname, '..', 'references', 'scenarios.mjs')).href);

const DEFAULT_BASE = 'https://api.deepseek.com/chat/completions';
const DEFAULT_MODEL = 'deepseek-chat';
const TEMP = 0.7;
const MAX_TOKENS = 8192;
Confidence
88% confidence
Finding
The script sends user-provided document content to an external DeepSeek API endpoint, which is a real data egress path. In a banking-writing context, inputs may contain sensitive internal, customer, financial, or compliance data, so transmitting them off-host without strong validation, minimization, and explicit consent can create confidentiality and regulatory risk.

Static analysis

Detected: suspicious.env_credential_access, suspicious.exposed_secret_literal

Environment variable access combined with network send.

Critical
Code
suspicious.env_credential_access
Location
scripts/bankai_write.mjs:96

File appears to expose a hardcoded API secret or token.

Critical
Code
suspicious.exposed_secret_literal
Location
SKILL.md:38