The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: Automated security scanners identified a malicious file within the skill's research directory.
Evidence: The file references/research/05-ai-tech-niche.md was flagged as FileRepMalware [Misc]. While the file appears to be a markdown document, the direct detection by reputation scanners indicates a severe security risk within the package.
Evidence: Installation via npx skills add alchaincyf/x-mentor-skill executes remote code from the npm registry. While common for skill deployment, this combines with the existing malware flag to elevate risk.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill automates the extraction and local storage of user-specific data from the web.
Evidence: Scenario E (Account Diagnosis) instructs the agent to scrape up to 100 tweets from a specific X.com profile and save the results, including interaction counts and timestamps, into the user-data/{username}/ directory.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data from X.com as its primary input for analysis, creating a high-risk surface.
Ingestion points: The skill uses computer-use or browser tools to navigate to public X profiles and capture tweet content (SKILL.md, Scenario E).
Boundary markers: Absent; there are no instructions for the agent to use delimiters or ignore instructions that may be embedded within the scraped tweet text.
Capability inventory: The agent has filesystem write access to the user-data/ directory and full browser navigation capabilities to interact with X.com.
Sanitization: Absent; there is no evidence of filtering, escaping, or validation of the ingested tweet data before it is processed by the model to generate reports.
[COMMAND_EXECUTION]: Extensive use of automated browser control and computer interaction tools.
Evidence: The skill explicitly directs the agent to use computer-use to navigate to x.com, take screenshots, and scroll through pages to accumulate data. This level of autonomy in a browser environment can be redirected to perform unintended actions if the agent encounters malicious content.

x-mastery-mentor