The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data files to generate scientific metrics, creating an attack surface where malicious instructions could be embedded in data. * Ingestion points: Processes CIF structure files from the inputs directory and metadata from the task_csv file, as described in SKILL.md and references/metrics.md. * Boundary markers: There are no specific instructions or delimiters mentioned to prevent the agent from following potential natural language instructions found within the data files. * Capability inventory: The skill possesses Bash execution, Read, and Write capabilities to run evaluation tools and process results, as defined in the YAML frontmatter of SKILL.md. * Sanitization: No sanitization or validation logic is specified for the contents of the ingested structure or CSV files.
[DYNAMIC_EXECUTION]: The skill includes a mechanism for extending functionality through custom Python code. As detailed in SKILL.md (Step 4) and references/metrics.md, the tool automatically loads and executes Python Metric subclasses placed in the src/rfproteina/metrics/custom_metrics/ directory. This dynamic loading of local scripts constitutes an execution surface.

rfd4-proteina-evaluate