The Agent Skills Directory

[PROMPT_INJECTION]: The instructions in references/core_workflow.md contain explicit directives to override the AI agent's default behavior and constraints. The text states that the skill's formatting requirements have "highest priority" and must be 100% followed even if they conflict with the agent's internal rules or constraints (e.g., "highest priority, covering all default behaviors", "forbidden to skip template requirements for any default rule").
[DATA_EXFILTRATION]: The script scripts/gzh_growth_fetcher.py accesses sensitive local file paths. Specifically, the _get_api_key function attempts to open and read user shell configuration files including ~/.zshrc, ~/.bashrc, ~/.bash_profile, and ~/.profile using regular expressions to extract API keys. Accessing these files is a security risk as they may contain other sensitive environment variables or credentials.
[COMMAND_EXECUTION]: The skill's core functionality relies on executing a local Python script (scripts/gzh_growth_fetcher.py) via the shell to fetch and process data from the RedFox API.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests untrusted data from an external API and displays it to the user.
Ingestion points: The skill retrieves account nicknames and links from the redfox.hk API endpoint in scripts/gzh_growth_fetcher.py.
Boundary markers: No delimiters or "ignore embedded instructions" warnings are used when interpolating the API data into the output templates defined in references/core_workflow.md.
Capability inventory: The skill possesses the ability to execute local scripts and perform network operations.
Sanitization: The skill does not perform any validation or escaping of the nicknames or links returned by the API before including them in the final markdown output.

douyin-rise-ranking