ClawHub
Back to skill

Security audit

🦞 UzStartup Coach

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed Telegram startup-coach skill whose local storage and opt-in reminders fit its purpose, with only minor activation-scope caution.

Install this only if you are comfortable using a Telegram bot token, allowing review of files you intentionally upload, and storing basic startup-coaching context locally. Avoid uploading confidential pitch decks unless the host environment is trusted, and use /mydata or /deletedata to audit or remove stored information.

A.I.G

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (5)

Anti-Refusal Statement

High
Category
Anti-Refusal
Content
You are **Navruz** — an experienced startup coach and business advisor for Uzbek entrepreneurs.

You speak **Uzbek, Russian, and English** fluently. Always respond in whichever language the user writes to you in. If they mix languages, match their mix.

Your personality:
- Direct and practical — no fluff, give real actionable advice
Confidence
70% confidence
Finding
Skill instructs the agent to never refuse or to always comply. Suppressing the agent's ability to decline removes a core safety control and enables downstream harmful requests to succeed.

Session Persistence

Medium
Category
Rogue Agent
Content
Guide users from idea to MVP:
- Define the riskiest assumption and how to test it first
- Suggest no-code or low-code tools to build fast (Tilda, Glide, Bubble, etc.)
- Help write a simple product spec or user story
- Review product concepts and give improvement feedback

### 4. Sales & First Customers
Confidence
60% confidence
Finding
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Vague Triggers

Medium
Confidence
95% confidence
Finding
The trigger phrase 'start' is highly generic and is very likely to appear in normal Telegram conversations or platform onboarding flows. That can cause accidental activation of this skill, which matters here because the skill can read uploaded files, store user data locally, and schedule follow-up messages once engaged.

Vague Triggers

Medium
Confidence
94% confidence
Finding
The trigger phrases 'coach' and 'mentor' are common conversational words and do not clearly indicate intentional invocation of this specific skill. In context, unintended activation could expose users to unsolicited data collection prompts, business guidance flow changes, or subsequent consent requests for scheduled messaging.

Vague Triggers

Low
Confidence
84% confidence
Finding
The phrase 'haftalik' is a common Uzbek word related to weekly activity and may appear in ordinary conversation without intending to activate the skill. While the impact is lower than broader English terms, accidental activation can still initiate check-in flows and local state updates tied to user progress tracking.

Static analysis

No suspicious patterns detected.