The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses Rscript -e to dynamically generate and execute R code for extracting numeric results from .rds files during the result extraction process.\n- [COMMAND_EXECUTION]: The skill employs the Bash tool to trigger analysis pipelines (e.g., Rscript scripts/R/00_run_all.R) and perform automated data extraction from tables and logs using regex patterns.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from research manuscripts (.tex, .md, .pdf) and output logs. It lacks boundary markers or sanitization when processing this data, which could be used to influence extraction logic via hidden instructions.\n
Ingestion points: Manuscript paths ($0), output directories ($1), and log files scanned in Phases 1 and 2.\n
Boundary markers: No explicit delimiters are defined for parsing manuscript content.\n
Capability inventory: Includes Bash for shell command execution, Write for generating quality reports, and Monitor for managing background tasks.\n
Sanitization: No explicit filtering or escaping of manuscript text is implemented before processing.\n- [DATA_EXFILTRATION]: The skill reads manuscript contents and environment captures (e.g., sessionInfo.txt), bringing potentially sensitive research data and system metadata into the agent context.

audit-reproducibility