The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill connects to api.deepseek.com to perform its core function. This is an official endpoint for a well-known AI service and is documented as the intended destination for API requests.
[COMMAND_EXECUTION]: The skill executes a local script at scripts/chat.mjs using the Bash tool. This is standard behavior for skill harnesses and does not involve the execution of arbitrary or untrusted code.
[INDIRECT_PROMPT_INJECTION]: The skill takes user-provided prompts and system messages as arguments. This is an inherent attack surface for LLM tools; however, the skill serves as a direct interface and does not include risky administrative or destructive capabilities. Ingestion points: --prompt and --system arguments. Boundary markers: None identified in the wrapper. Capability inventory: Bash execution for local scripts and network POST requests to the DeepSeek API. Sanitization: No input sanitization is mentioned.

deepseek-chat