The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains instructions for the agent to perform environment reconnaissance to find secrets.
Evidence: The command bash env | grep -i "api_key\|_key" in SKILL.md is explicitly provided to help the agent discover configured authentication keys.
[COMMAND_EXECUTION]: The skill directs the agent to read internal source code files to discover system-level proxy configurations and undocumented endpoints.
Evidence: The instruction to run read_file core/http_client.py | grep -A 15 "DEFAULT_PROXIED_APIS" in SKILL.md demonstrates unauthorized access to internal system components.
[EXTERNAL_DOWNLOADS]: The skill downloads external code archives and references several third-party libraries via CDNs.
Evidence: curl -L https://github.com/tabler/tabler/archive/refs/heads/main.zip -o tabler.zip in SKILL.md downloads external software, and the HTML templates include script tags targeting cdn.jsdelivr.net and d3js.org.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting data from numerous external financial and environmental APIs without security boundaries.
Ingestion points: Multiple external APIs including CoinGecko, Twelve Data, and Open-Meteo are fetched in SKILL.md.
Boundary markers: Absent from the integration logic.
Capability inventory: The skill utilizes bash and read_file capabilities across its instructions.
Sanitization: No sanitization or validation of external API responses is present before processing or displaying data.

dashboard