The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions for installing dependencies like the Grok CLI by downloading and executing shell scripts from official domains using piped shell commands.
[DATA_EXFILTRATION]: The skill contains logic to extract and decrypt sensitive authentication data from the local environment, including session cookies from Chrome, Firefox, and Safari, and encryption keys from the macOS Keychain. This information is used to facilitate authenticated research requests.
[COMMAND_EXECUTION]: The system relies on the Python subprocess module to execute various CLI tools and system commands, such as openssl for decryption and yt-dlp for YouTube processing, as part of its multi-backend orchestration.
[EXTERNAL_DOWNLOADS]: The skill interacts with numerous external APIs and services, including ScrapeCreators, Valyu AI, Brave Search, and Exa AI, and utilizes the @lev-os/timetravel library provided by the skill author.
[INDIRECT_PROMPT_INJECTION]: By ingesting and summarizing untrusted data from social media and search results, the skill is exposed to potential indirect prompt injection attacks where malicious instructions in the source material could influence the agent's output.
[CREDENTIALS_UNSAFE]: The skill manages multiple API keys through environment variables and local configuration. While it follows standard secret management practices, the automated harvesting of browser session tokens is a high-privilege operation.

research