The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to search for and execute canonical analysis scripts such as analysis.R, run_*.py, find_*.R, and *.Rmd found within the data folder 'as-is'. This behavior allows the execution of arbitrary code that may be present in a potentially untrusted dataset.
Evidence: SKILL.md contains 'Canonical analysis scripts (analysis.R, run_.py, find_.R, *.Rmd) → execute as-is and read the output.'
[DYNAMIC_EXECUTION]: The skill relies on the agent to 'write and run Python code via Bash' to perform statistical calculations and data processing. The CODE_REFERENCE.md provides numerous templates for runtime code generation and execution involving pandas, scipy, and networkx.
Evidence: SKILL.md states 'When analysis requires computation... write and run Python code via Bash. Don't describe what you would do — execute it and report actual results.'
[INDIRECT_PROMPT_INJECTION]: The skill ingests content from external data sources including mass-spec outputs (CSV, TSV), pre-computed results, and executed notebooks. These ingestion points lack explicit boundary markers or sanitization, potentially allowing malicious instructions embedded in data files to influence agent behavior.
Ingestion points: reads *_executed.ipynb, proteinGroups.txt, *_Report.tsv, and various result CSVs.
Boundary markers: Absent; no instructions are provided to the agent to ignore instructions embedded in the data files.
Capability inventory: Full access to Bash/Python execution and network access to scientific databases.
Sanitization: Not mentioned; the skill uses standard library functions to load data directly into analysis pipelines.

tooluniverse-proteomics-analysis