The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches its core logic and orchestration scripts from a third-party GitHub repository (github.com/yuwanpai2004-create/codex-skills) which is not an established trusted source.
[REMOTE_CODE_EXECUTION]: The installation instructions require the user to run a bootstrap script (python scripts/bootstrap.py install) that is downloaded from the external repository. This pattern of downloading and executing remote code poses a security risk if the repository is compromised.
[COMMAND_EXECUTION]: The skill's operation relies on the execution of several Python scripts (init_profile_map.py, harvest_social_corpus.py) that perform file system operations and network requests to external social media platforms.
[INDIRECT_PROMPT_INJECTION]: The primary function of the skill is to harvest public data from social media platforms like Xiaohongshu and Douyin. This untrusted data could contain malicious instructions designed to influence the behavior of an AI agent if the dataset is later used for RAG (Retrieval-Augmented Generation) or analysis. Ingestion points: scripts/harvest_social_corpus.py writes external content to the corpus/ directory. Boundary markers: None documented; the skill stores content in Markdown format without explicit security delimiters. Capability inventory: The skill possesses the capability to execute shell commands, perform network requests, and modify local files via the provided Python scripts. Sanitization: No sanitization or filtering of the harvested social media content is mentioned, allowing raw external text to enter the agent's environment.

Social Corpus Harvest