The Agent Skills Directory

Remote Code Execution (HIGH): The skill facilitates the download and direct execution of a shell script from an untrusted external domain (get.nextflow.io) using the piped bash pattern. This pattern allows for arbitrary code execution without verification of the script content and is flagged as a high-risk security issue.\n
Evidence: Found in SKILL.md and references/installation.md: curl -s https://get.nextflow.io | bash.\n- Command Execution (HIGH): The skill instructs the agent to execute commands with elevated privileges (sudo) for environment setup and service management, which can lead to unauthorized system modifications if misused.\n
Evidence: SKILL.md contains sudo usermod -aG docker $USER and sudo systemctl start docker.\n- External Downloads (MEDIUM): The skill is designed to interact with and download data from external genomic repositories (NCBI/GEO/SRA), involving significant network operations to non-whitelisted domains.\n
Evidence: Referenced in scripts/utils/__init__.py and Step 0 of SKILL.md.\n- Indirect Prompt Injection (LOW): The skill processes user-supplied file and directory names to infer metadata for pipeline configuration. These values are used in command-line templates, creating a surface for indirect prompt injection.\n
Ingestion points: scripts/detect_data_type.py and scripts/utils/file_discovery.py scan user-provided directories.\n
Boundary markers: None identified in command templates.\n
Capability inventory: Arbitrary shell command execution via nextflow and local python scripts.\n
Sanitization: Partial sanitization via regex in scripts/utils/sample_inference.py.

nextflow-development