The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied disease names to generate research protocols and code, creating a potential surface for indirect prompt injection.
Ingestion points: User input for the disease pair phenotyoes is used throughout the SKILL.md workflow to customize the output.
Boundary markers: The instructions do not include specific delimiters or 'ignore embedded instructions' warnings for the data provided by the user.
Capability inventory: The skill has the capability to generate executable R code, figure plans, and complex study protocols.
Sanitization: There is no explicit sanitization or validation mentioned for the disease names before they are interpolated into the plan or code blocks.
[EXTERNAL_DOWNLOADS]: The skill facilitates the acquisition of data and software from trusted scientific and technical repositories.
Data Acquisition: Instructions include fetching data from the NCBI Gene Expression Omnibus (GEO) via the GEOquery package.
Software Installation: The skill provides commands for installing R packages from CRAN and Bioconductor registries.
Vendor Resources: The skill references the author's official repository at github.com/aipoch/medical-research-skills for medical research skills and documentation.
[METADATA_POISONING]: The skill includes an external metadata file (eval_report_dual-disease-transcriptomic-ml-planner_polished_result.json) containing self-authoritative safety claims.
Evidence: The file includes a safety verdict of 'PASS' and claims of scientific integrity. These self-claims are noted as metadata but are not used to influence the analyzer's independent verdict.

dual-disease-transcriptomic-ml-planner