The Agent Skills Directory

[DATA_EXPOSURE]: The scripts scripts/fetch_topic_aweme_list.py and scripts/search_tiktok_topics.py are designed to read from sensitive shell configuration files, specifically ~/.zshrc, ~/.bashrc, ~/.bash_profile, and ~/.profile. While the intent is to find the REDFOX_API_KEY, these files often contain other sensitive credentials, API keys, and private configuration data that should not be broadly accessed by individual skills.
[DATA_EXPOSURE]: Both scripts implement a custom network requester, post_redfox_no_sni, which explicitly disables SSL hostname verification by setting context.check_hostname = False. This bypasses a fundamental security mechanism of HTTPS, making the connection to the redfox.hk API vulnerable to Man-in-the-Middle (MitM) attacks where an attacker could intercept or modify data in transit.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content retrieved from external TikTok topic and work lists, which presents an attack surface for indirect prompt injection.
Ingestion points: Data is fetched from the topicList and workList fields via the RedFox API and displayed to the user.
Boundary markers: The skill uses Markdown tables to present the data, which provides some structural separation but does not prevent command injection if the LLM interprets the table content as instructions.
Capability inventory: The skill possesses network access (via raw socket and ssl modules) and file-read capabilities (accessing shell profiles).
Sanitization: The scripts include an escape_md function that escapes pipe characters and removes newlines to prevent simple Markdown table breakage, though this is not a comprehensive defense against LLM-based prompt injection.

tiktok-keyword-topic-search