The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a checklist and workflow section in SKILL.md that contains shell commands for code linting (ruff check), local testing of Python modules (python -m tooluniverse.cli), and Git repository management (git, gh). These are standard operations for a development-focused agent working within a codebase.
[DYNAMIC_EXECUTION]: The references/code-patterns.md file documents a pattern for executing R scripts by writing them to temporary files and running them via subprocess.run. This is presented as a best-practice mechanism to resolve shell-escaping issues in R subprocess calls.
[INDIRECT_PROMPT_INJECTION]: The skill outlines implementation patterns for tools that consume data from external bioinformatics APIs (e.g., CIViC, GtoPdb, GTEx). It identifies risks such as path injection in API endpoints and explicitly provides a mitigation pattern using regex validation (re.fullmatch(r"[A-Za-z0-9._-]+", value)) to sanitize user-provided parameters before they are interpolated into URLs.
Ingestion points: Data returned from bioinformatics APIs like CIViC, GtoPdb, and ENCODE (specified in references/api-fixes.md).
Boundary markers: The skill recommends providing a normalization_note or hint in the tool output to disclose automated data transformations to the user.
Capability inventory: The skill patterns utilize subprocess.run for R script execution and requests.get for network operations.
Sanitization: The skill provides an explicit regex-based sanitization pattern for path parameters in the NVIDIA NIM integration section to prevent path injection.

devtu-code-optimization