The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied parameters such as strategy names, symbols, and date ranges and interpolates them directly into shell commands. * Ingestion points: User input for <strategy-or-model>, <TICKER>, and <period> in SKILL.md. * Boundary markers: The skill does not use delimiters or instructions to ignore embedded commands within these inputs. * Capability inventory: The skill uses managed_agent_prompt and Bash, allowing for shell execution in the remote container. * Sanitization: There is no evidence of input validation or escaping before interpolation into the message string.
[COMMAND_EXECUTION]: The skill builds and dispatches shell commands (e.g., npx neural-trader --backtest --strategy <name> --symbol <TICKER>...) to a remote managed agent. An attacker providing malicious input for the strategy or symbol parameters could potentially execute arbitrary commands within the cloud-hosted environment.
[EXTERNAL_DOWNLOADS]: The skill automatically installs the neural-trader package from the NPM registry during the container provisioning phase. The implementation follows security best practices by using the --ignore-scripts flag to prevent the execution of potentially malicious lifecycle scripts during installation.

trader-cloud-backtest