The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill utilizes npx to fetch and run code from a specific commit in a third-party GitHub repository (socai-io/jev-social). This execution of unverified remote code poses a significant security risk.
Evidence: The skill defines several commands using npx github:socai-io/jev-social#baf3cd6aa4f9c881665c29ed29a10391f761760b for status checks, onboarding, research actions, and starting a local server.
[EXTERNAL_DOWNLOADS]: The skill depends on external software components, including the jev-social package from GitHub and the socai CLI tool. It also performs network operations to social media platforms and the OpenRouter API.
[COMMAND_EXECUTION]: The skill uses npx to execute shell commands for interacting with its research tools and managing local server processes.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from untrusted social media platforms (Instagram, TikTok, LinkedIn), which is a common vector for prompt injection attacks.
Ingestion points: Data is collected from social media posts, comments, and profiles via the socai tool as described in SKILL.md.
Boundary markers: The instructions explicitly warn the AI to treat social media content and tool output as evidence, not as instructions, to prevent obedience to embedded commands.
Capability inventory: The skill can execute shell commands via npx, write files to the local system, and communicate with external APIs.
Sanitization: The skill documentation mentions the use of sanitized evidence and checkpoints to reduce the risk of processing malicious content.

jev-social