The Agent Skills Directory

[COMMAND_EXECUTION]: Local diagnostic scripts like doctor_yichen.py and validate_family.py use subprocess.run to check the status of installed CLI tools (e.g., gh, yt-dlp). These calls are limited to metadata commands and employ strict environment variable filtering to protect user secrets.
[PROMPT_INJECTION]: The skill handles data from the web, creating a potential indirect prompt injection surface. This is mitigated by a mandatory research protocol that requires verifying all findings against original sources, preventing the agent from following instructions hidden in external data.
[CREDENTIALS_UNSAFE]: The skill refers to various API keys and tokens required for its backends. It correctly manages these via environment variables and includes safety checks that confirm their presence without reading or exposing the secret values.
[REMOTE_CODE_EXECUTION]: The skill uses external CLIs for data retrieval but does not perform automated installation or execution of untrusted remote code. It provides utilities for users to verify their local tool configuration safely.
[EXTERNAL_DOWNLOADS]: The skill references several external services and tools (e.g., Firecrawl, GitHub, YouTube-DLP) as research backends. These are recognized as well-known services, and the skill only invokes them if they are already present in the user environment.

yichen-web-research