Kaiji Fitness Coach

• Outcome: Pass
• Latest audit: Jun 9, 2026 at 8:03 AM UTC
• Version:
  2.1.0

Overview

This is a fitness coaching skill with expected local setup and personalization behavior, with no evidence of hidden data exfiltration or destructive behavior outside its own exercise database folder.

Install if you are comfortable with a fitness skill asking about body metrics, injuries, goals, and training history. Run the database setup only from the installed skill directory, and use --force only when you intentionally want to replace the local exercise database.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (6)

Vague Triggers

Medium

• Confidence: 83% confidence
• Finding: The trigger phrases are very broad and overlap with ordinary fitness conversation, making accidental invocation likely. Overbroad activation increases the chance that the skill is engaged in contexts where the user did not intend data analysis, reference loading, or any connected tooling, which can expose unnecessary functionality and degrade user control.

Vague Triggers

Medium

• Confidence: 90% confidence
• Finding: The trigger phrases are broad natural-language requests such as asking for training advice or report analysis, which can overlap with ordinary conversation and cause unintended skill activation. In this fitness context, that mainly risks incorrect routing, surprise behavior, or analysis of user-provided data when the user did not explicitly intend to invoke this specific skill.

Vague Triggers

Medium

• Confidence: 95% confidence
• Finding: The trigger conditions are broad enough to match many ordinary fitness-related requests, which can cause the skill to activate when the user did not clearly request this specific capability. Over-broad activation increases the chance of unintended instruction injection into unrelated conversations and can interfere with routing, especially because the skill is designed to provide authoritative coaching guidance.

Vague Triggers

Medium

• Confidence: 89% confidence
• Finding: The trigger phrases are very broad, everyday nutrition questions like '吃什么好' and '怎么吃能增肌', which can cause the skill to activate in contexts the user did not explicitly intend. In a health-adjacent skill, unintended activation increases the chance of users receiving generic dietary guidance without sufficient personalization, which can be unsafe for people with medical conditions or special populations.

Missing User Warnings

Medium

• Confidence: 96% confidence
• Finding: The skill instructs collection and structured storage of personal profile data including age, weight, height, injuries, and lifestyle-related details, but provides no notice, consent flow, retention limits, or privacy safeguards. In a fitness context this is more sensitive because injury and body metrics can reveal health-related information, increasing privacy and misuse risk if stored or reused across sessions.

Missing User Warnings

Medium

• Confidence: 91% confidence
• Finding: When `--force` is used, the script recursively deletes the existing database directory with `shutil.rmtree` and does so without confirmation, backup, or additional path-safety checks. If the target path is wrong, replaced, or unexpectedly linked in a compromised environment, this can cause destructive local data loss.

VirusTotal

41/41 vendors flagged this skill as clean.