The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external research data and literature metadata which represents a standard injection surface for agents ingesting external content.
Ingestion points: The workflow ingests data from TSV files (conformer_rmsf.tsv, binder_hotspot_matrix.tsv, literature_cases.tsv, assay_cascade.tsv, evidence_table.tsv) and PDB/mmCIF structural files.
Boundary markers: The skill enforces 'declarative data contracts' (strict schemas) for all input data and utilizes deterministic geometry gates to audit the resulting figures.
Capability inventory: The skill utilizes Bash for environment setup, Read/Write for file operations, and executes Python plotting and audit scripts.
Sanitization: Scientific findings are validated against primary records (NCBI E-Utilities) and the final output is reviewed by a 'Vision Model Semantic Review' using focused, single-question queries to mitigate hallucination or instruction following from data content.
[COMMAND_EXECUTION]: The skill executes shell commands to configure a project-local virtual environment using uv and to run the bundled audit and plotting scripts. This activity is restricted to the local workspace and intended for the primary function of the skill.
[DYNAMIC_EXECUTION]: The production workflow involves the agent writing Python plotting scripts (e.g., fig1.py, render_script.py) based on domain-specific reference guides and then executing them via the local interpreter. This is a controlled script generation process intended for scientific visualization.

bioresearcher-plot-making