The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill exhibits a high-risk attack surface where data from the FastFold API (api.fastfold.ai) is used to influence agent behavior and perform file operations.
Ingestion points: API responses are ingested in scripts/download_cif.py, scripts/fetch_results.py, and scripts/wait_for_completion.py via the get_results function.
Boundary markers: There are no delimiters or instructions to ignore embedded commands when processing API JSON.
Capability inventory: The skill can perform network requests (requests.get/post) and write files to the local disk (scripts/download_cif.py).
Sanitization: No validation or sanitization is performed on API-provided data, such as cif_url or job_id, before use in side-effecting operations.
External Downloads (HIGH): scripts/download_cif.py downloads content from URLs provided in the API response and writes them to the local filesystem. If the API is compromised or manipulated, it could be used to deliver malicious payloads or overwrite system files.
Path Traversal (MEDIUM): In scripts/download_cif.py, the job_id argument is directly concatenated into a file path: os.path.join(args.dir, f"{args.job_id}.cif"). A malicious job ID containing ../ could allow writing files outside the intended directory.
Unverifiable Dependencies (LOW): The scripts require the requests library. While a standard package, it is an external dependency that must be installed by the user and is not from a whitelisted trusted source.

fastfold-fold-job