The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided protein structure files (PDBs) which can serve as an injection vector if they contain malicious metadata or comments designed to manipulate agent behavior during the design and validation cycle.
Ingestion points: Processes PDB files via the inference.input_pdb parameter in SKILL.md.
Boundary markers: Absent; there are no instructions for the agent to sanitize or ignore non-structural content within the PDB files.
Capability inventory: The skill can execute shell commands (python, uv), perform file system operations (Read, Write, Edit), and dispatch tasks to remote compute services (alterlab-remote-compute).
Sanitization: Absent; the skill passes structure data directly to the inference script without validation of non-structural metadata.
[COMMAND_EXECUTION]: The skill uses shell commands to drive the RFdiffusion inference process.
Invokes run_inference.py using python or uv run to perform backbone generation and motif scaffolding.
[EXTERNAL_DOWNLOADS]: The skill utilizes external code and model weights from a recognized academic repository.
Fetches the RFdiffusion codebase from the official RosettaCommons GitHub repository.
Downloads pre-trained model weights (several GB) required for the diffusion process as part of the tool's installation and runtime setup.

alterlab-rfdiffusion