The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/spa_workflow.py uses subprocess.run to execute external RELION binaries such as relion_refine_mpi and relion_postprocess, as well as MPI runners like mpirun. While the script uses a list-based argument structure to prevent shell injection and includes a version check for the executable, the command names and paths are configurable via command-line arguments, creating a surface for executing arbitrary local binaries if an agent is misdirected.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from STAR and MRC files to configure refinement parameters and validate stacks, which represents an attack surface for data-driven influence on the agent's actions.
Ingestion points: STAR files are parsed using the starfile library in validate_star, and MRC headers/data are read in read_map and check_halves within scripts/spa_workflow.py.
Boundary markers: None; the script processes the file content directly for validation and command construction without explicit data delimiters in the prompt interpolation.
Capability inventory: The skill can execute shell commands via subprocess.run and write logs or processed data to the filesystem.
Sanitization: The script performs various validation checks, including regex matching for symmetry strings and image names, as well as range checks for physical parameters (e.g., voltage and defocus), which reduces the risk but does not fully eliminate the potential for maliciously crafted data to influence the processing pipeline.

relion