The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The configuration file references/investigator-config.json contains a hardcoded default API key (ak_b45b6a6881f4400fb321428947eb6661) for accessing the vendor's services.
[COMMAND_EXECUTION]: The script assets/daily_report.py executes shell commands to establish persistence, using launchctl on macOS and modifying the user's crontab on Linux/Windows systems.
[COMMAND_EXECUTION]: The main script utilizes subprocess.run with shell=True when managing crontab entries, which can be sensitive to command injection if paths or inputs are improperly handled.
[COMMAND_EXECUTION]: The script reads sensitive local environment configuration files, such as ~/.zshrc and ~/.bashrc, to extract API keys for authentication.
[EXTERNAL_DOWNLOADS]: The skill fetches dynamic video metadata from the vendor's API endpoint at https://redfox.hk/story/api/parseWork/queryBiliAiMsgs.
[COMMAND_EXECUTION]: Indirect Prompt Injection Surface:
Ingestion points: External video titles and tags are fetched from the Bilibili API and processed in assets/daily_report.py.
Boundary markers: Absent; external data is integrated into the analysis flow without specific delimiters or isolation instructions.
Capability inventory: The skill performs subprocess execution and file system operations.
Sanitization: There is no evidence of filtering or escaping untrusted external text (titles/tags) before it is analyzed by the agent, potentially allowing embedded instructions to influence agent behavior.

bili-ai-feed