The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill clones a repository from an unverified GitHub user (https://github.com/dauparas/ProteinMPNN.git) to access the required tools and models.
[REMOTE_CODE_EXECUTION]: The instructions involve executing a Python script (protein_mpnn_run.py) directly from the cloned repository, which constitutes execution of remote code.
[COMMAND_EXECUTION]: The skill uses shell commands (pip, git, and python) to install dependencies, retrieve the codebase, and execute the model.
[PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by processing external data files (PDB and JSONL) and using the resulting context to influence shell command execution.
Ingestion points: The skill reads protein structure files via --pdb_path and configuration files via --fixed_positions_jsonl as seen in SKILL.md.
Boundary markers: No specific delimiters or instructions are provided to the agent to isolate or ignore potential malicious content within these data files.
Capability inventory: The skill performs shell command execution (pip, git, python) as described in the Running it section of SKILL.md.
Sanitization: There is no evidence of sanitization or validation of the input file content before it is used in the workflow.

proteinmpnn