The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it interpolates untrusted user data into LLM prompts. While the impact is limited to the context of the negotiation session, an attacker could provide crafted intent strings to influence the agent's bargaining behavior.
Ingestion points: The buyer_intent and seller_capability parameters in the run_negotiation method (scripts/negotiation_engine.py) accept arbitrary text from users.
Boundary markers: Absent. The inputs are directly embedded into the system prompts without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill's capabilities are restricted to network communication with the DashScope API via httpx. No file system write access or shell command execution is present.
Sanitization: Absent. There is no evidence of input validation, escaping, or filtering of the user-provided strings before interpolation.
[EXTERNAL_DOWNLOADS]: The skill integrates with external technological services and libraries to perform its core functions.
Dependencies: The skill requires standard Python packages httpx, python-dotenv, and rich from the official PyPI registry.
Well-known services: The NegotiationEngine (scripts/negotiation_engine.py) communicates with Alibaba Cloud's DashScope API (dashscope.aliyuncs.com) for Qwen model inference. This is a recognized enterprise service provider.

agent-negotiation-protocol