The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct users to run npx skills update rlm (or equivalents like pnpx or bunx) to check for and install updates. This facilitates the execution of scripts from an external package manager.
[EXTERNAL_DOWNLOADS]: The skill documentation specifies several external dependencies, such as predict-rlm, DSPy, Deno, and beautifulsoup4. These are fetched from official public registries during setup and usage.
[DYNAMIC_EXECUTION]: The primary feature of the skill is its architecture, which allows a Large Language Model (LM) to generate and execute Python code dynamically. This execution occurs within a sandboxed REPL environment (using Pyodide and Deno) to provide isolation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process potentially untrusted external content, such as documents and criteria strings, creating a surface for indirect prompt injection.
Ingestion points: External data enters the context through documents (File list) and criteria (CtxStr) fields in the DSPy signature.
Boundary markers: The provided examples do not demonstrate the use of explicit delimiters or instructions to ignore embedded commands within the processed data.
Capability inventory: The skill can execute Python code, call host-side tools, and access authorized network domains.
Sanitization: There is no evidence of pre-processing or sanitization of input data before it is presented to the LLM for interpretation and code generation.

rlm