alphafold-skill
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [Arbitrary File Write]: The script
scripts/rest_request.pyincludes asave_rawfeature that allows the agent to write the API response to a file path specified by theraw_output_pathinput field. While intended for saving large JSON payloads, this capability could potentially be used to write data to sensitive locations on the filesystem if the execution environment does not enforce strict path sandboxing. - [Generic Network Client]: Although the skill is designed for the AlphaFold Protein Structure Database, the underlying Python script is a generic REST client. It accepts any
base_url,method, andheadersprovided in the input JSON. This flexibility could allow the skill to be used to interact with internal or external services beyond its stated purpose if the agent's instructions are bypassed. - [Indirect Prompt Injection Surface]: The skill processes data from external REST APIs and returns summaries or raw JSON to the agent. This represents an indirect prompt injection surface where a compromised or malicious API endpoint could return instructions designed to influence the agent's behavior.
- Ingestion points: Data is ingested from external URLs via the
requestslibrary inscripts/rest_request.py. - Boundary markers: None detected; the skill truncates output for length but does not use specific delimiters to isolate external content from instructions.
- Capability inventory: The skill can perform network requests (
requests.Session().request) and write files to the local disk (Path.write_text). - Sanitization: The script performs basic type validation for inputs but does not sanitize the content of the API response before it is returned to the agent context.
Audit Metadata