Security audit

KWDB Troubleshooting

Security checks for vulnerabilities and agentic risk

Overview

This skill is a coherent KWDB troubleshooting guide, with expected log and SQL handling risks that users should manage through redaction and access control.

Install only for teams comfortable with Chinese diagnostic output. When using it on real incidents, treat logs, SQL text, object names, process arguments, and metrics as sensitive; redact secrets and customer-specific details before pasting them into chats, tickets, or generated reports.

SkillSpector

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep

Findings (6)

Natural-Language Policy Violations

Medium

Confidence: 89% confidence
Finding: The skill hard-codes the response language to Chinese regardless of the user's preference or platform expectations. This can reduce transparency and operator comprehension during incident handling, especially in multilingual environments, and may cause analysts to miss or misinterpret diagnostic results rather than creating a direct system-compromise path.

Natural-Language Policy Violations

Medium

Confidence: 94% confidence
Finding: The rule `always reply in Chinese` hard-codes output language and removes user choice or policy-driven language selection. This is not code-execution dangerous, but it can cause policy noncompliance, reduce accessibility, and interfere with downstream systems or users expecting another language during incident handling.

Natural-Language Policy Violations

Medium

Confidence: 92% confidence
Finding: The file hard-requires Chinese output for Mode 1 regardless of the user's preference or language, which is a policy and usability control issue. In a troubleshooting skill, this can cause misunderstandings, reduce operator effectiveness, and make incident handling harder when responders expect another language, though it is not code-execution or data-exfiltration behavior.

Natural-Language Policy Violations

Medium

Confidence: 92% confidence
Finding: The seven-section mode also mandates Chinese output without user choice, creating the same language-lock behavior in a structured incident-report workflow. In operational and forensic contexts, forced language constraints can impair collaboration, introduce translation errors, and reduce the utility of generated reports for mixed-language teams.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The playbook explicitly directs responders to capture slow SQL text during incident triage, but it provides no guardrails about secrets, personal data, tenant identifiers, or literals that may be embedded in SQL statements. In an incident workflow, that omission can cause sensitive query text to be copied into tickets, chats, or reports and broaden data exposure beyond the production system.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The instruction to include object names, SQL text, or operations in the report can leak sensitive business context, customer identifiers, schema details, or embedded literals into downstream artifacts. Because this is a troubleshooting skill, reports are likely to be shared widely during incidents, which increases the chance of unnecessary disclosure.

Static analysis

No suspicious patterns detected.