The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's README.md provides instructions to download strategy scripts and configuration files from a remote GitHub repository (https://raw.githubusercontent.com/Senpi-ai/senpi-skills/). These files are necessary for the skill's primary functionality.
[REMOTE_CODE_EXECUTION]: The installation process involves fetching external Python scripts and executing them. Furthermore, the scripts/lemur_config.py file dynamically modifies sys.path by calculating the location of the senpi-trading-runtime dependency at runtime, which is a form of dynamic code loading.
[COMMAND_EXECUTION]: The README.md documentation includes shell commands for creating workspace directories, downloading external files, and executing the strategy producer as a background daemon using nohup and disown.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. Untrusted data is ingested from external market and sentiment APIs via MCP tools in scripts/lemur-producer.py. This data (e.g., asset names, trend descriptions, and trader sentiment) is interpolated directly into the LLM decision prompt in runtime.yaml using the {{signal_lemur_signals}} placeholder. There are no sanitization mechanisms or explicit boundary markers to prevent malicious content within this external data from influencing the LLM's logic for the OPEN_POSITION action.
[CREDENTIALS_UNSAFE]: The skill requires the user to provide a sensitive authentication token (SENPI_AUTH_TOKEN) through environment variables. This token is used by the producer scripts to authenticate requests to MCP tools and signal processing services.

lemur-strategy