The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches the neural-trader package from the NPM registry using the npx command. This downloads code from a remote repository to the local environment at execution time.
[REMOTE_CODE_EXECUTION]: The use of npx neural-trader@latest results in the execution of external code. The package includes native NAPI functions and SIMD-optimized binaries, which represent a complex execution surface. The use of the @latest version tag means the exact code version is not pinned and could change between runs.
[COMMAND_EXECUTION]: The skill heavily utilizes shell commands to manage trading operations, including initializing workspaces and running backtests.
[PROMPT_INJECTION]: The skill ingests untrusted data from market data files (--data) and strategy files (--strategy), creating a surface for indirect prompt injection.
Ingestion points: Files processed by the train, backtest, and run commands in SKILL.md and references/commands.md.
Boundary markers: No delimiters or warnings for embedded instructions are present.
Capability inventory: The agent can execute shell commands via npx and access the local file system.
Sanitization: No sanitization of the content within the data or strategy files is specified.

neural-trader