The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill pulls multiple Docker images to establish its search infrastructure.
Fetches ghcr.io/flaresolverr/flaresolverr for Cloudflare challenge bypass.
Downloads ghcr.io/chenyme/grok2api and ghcr.io/xuncv/tavilyproxymanager to handle API aggregation and load balancing.
Connects to the well-known firecrawl.dev API as a fallback for web scraping operations.
[COMMAND_EXECUTION]: The skill provides a set of shell scripts (grok-search.sh, tavily-search.sh, web-fetch.sh, web-map.sh, dual-search.sh) that the agent executes via Bash to perform search and retrieval tasks.
[PROMPT_INJECTION]: Analysis of SKILL.md reveals specific behavioral instructions for the agent, such as prioritizing X/Twitter for certain queries and requiring cross-verification of facts. These are functional guidelines for the skill's operation and do not attempt to bypass agent safety filters.
[CREDENTIALS_UNSAFE]: The skill manages sensitive data like Grok SSO tokens and Tavily API keys. It uses a local .env file and a dedicated import-keys.sh script to securely load these into local services. The documentation explicitly warns the agent against echoing these sensitive values in the conversation context.
[DATA_EXPOSURE]: The skill is configured to bind its internal service ports (8100 and 8200) to 127.0.0.1 by default, preventing external access to the local API aggregation services.

ultimate-search