Karpathy Wiki Improve

• Outcome: Pass
• Latest audit: Jun 17, 2026 · 9:25 PM UTC
• Version: 1.1.2

Overview

This is a disclosed markdown-wiki helper that can write wiki files and perform web research, with some activation and automation wording users should treat carefully.

Install only if you want an agent to maintain a local markdown wiki. Configure the wiki root narrowly, review planned file changes before ingest/relink/research runs, and do not allow periodic maintenance unless you explicitly want automatic checks on that wiki.

SkillSpector

By NVIDIA

Vulnerability Patterns

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

Findings (6)

Vague Triggers

Medium

• Confidence: 87% confidence
• Finding: The example triggers are broad natural-language phrases such as 'research X', 'relink', and 'run lint on my wiki', which can overlap with ordinary user requests and cause the skill to activate when the user did not explicitly intend to invoke it. Because this skill performs ingest, web research, and knowledge-graph maintenance, unintended activation could lead to unnecessary external lookups or modifications to persistent wiki content.

Vague Triggers

Medium

• Confidence: 93% confidence
• Finding: The activation keywords are very broad and include common phrases like 'research X', 'wiki', and 'knowledge graph', which raises the likelihood that the skill will activate when the user did not explicitly intend to invoke a file-modifying knowledge-management workflow. Because this skill can create and update local files and later trigger additional actions, accidental invocation could lead to unintended local state changes and external fetch activity.

Vague Triggers

Medium

• Confidence: 90% confidence
• Finding: The skill specifies 'periodic heartbeat triggers lint + relink + quality check' without defining how consent, scheduling, or visibility are handled. That creates a risk of background or unsignaled execution that can modify wiki files over time without a fresh user request, which is unsafe for a local-content management skill.

Vague Triggers

Medium

• Confidence: 88% confidence
• Finding: The use-case section includes broad, everyday requests such as 'help me digest this link' and 'organize my collected content on XXX', which are common user intents that may not imply consent to create, update, and relink local files. In context, the danger is amplified because the skill is not read-only; it performs structured writes, logging, and potentially web-backed research.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The top-level description advertises ingest, relink, lint, and knowledge-graph maintenance but does not warn that the skill will create and modify local wiki files. Users may invoke it expecting analysis only, while the workflow actually performs persistent writes to multiple paths including pages, indexes, and logs.

Missing User Warnings

Medium

• Confidence: 94% confidence
• Finding: The Deep Research section instructs the agent to run external web searches, collect results, store them under raw/sources, and then execute the full ingest pipeline, but this behavior is not surfaced as a user warning in the skill description. This creates privacy and consent risks because user topics may be transmitted to external services and the fetched results are stored locally for future processing.

VirusTotal

62/62 vendors flagged this skill as clean.

Static analysis

No suspicious patterns detected.