The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to clone source code from a third-party GitHub repository (https://github.com/dauparas/ProteinMPNN.git) which is not managed by a verified organization.
[REMOTE_CODE_EXECUTION]: The instructions direct the agent to execute a Python script (protein_mpnn_run.py) directly from the cloned repository. Downloading and executing code from an unverified third-party source poses a significant security risk if the source repository is compromised.
[COMMAND_EXECUTION]: The skill builds shell commands using user-controllable arguments like --pdb_path, --pdb_path_chains, and --sampling_temp. Executing these via the shell without sanitization is a high-risk pattern.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it ingests untrusted user data and interpolates it into executable commands without proper safeguards.
Ingestion points: User-supplied PDB file paths, chain identifiers, and sampling temperature strings (SKILL.md, lines 35-38).
Boundary markers: Absent; the instructions do not specify how to delimit user input or warn the agent to ignore instructions embedded within the data.
Capability inventory: The skill utilizes git, pip, and python to perform network operations, package installation, and script execution.
Sanitization: Absent; there are no instructions for escaping, validating, or filtering user-provided parameters before they are used in shell commands.

proteinmpnn