pubchem-pug-skill
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- Arbitrary File System Write: The tool includes a
save_rawparameter that writes network response data to a path provided in the input.\n - Evidence: The
_save_raw_outputfunction inscripts/rest_request.pyusesPath(raw_output_path).write_textto save content directly to the filesystem.\n - Context: This enables the agent to store data locally; however, without path restrictions, it could potentially be used to write to or overwrite sensitive files if the input parameters are not carefully controlled.\n- Configurable Network Access: The implementation allows the agent to specify any
base_urlandpath, enabling communication with arbitrary web endpoints.\n - Evidence: The
parse_inputandexecutefunctions inscripts/rest_request.pydo not restrict the target host or protocol, and_build_urlallows absolute URLs to override the base configuration.\n - Context: While designed for chemical data retrieval, this flexibility allows the tool to be used to reach other network resources if prompted.\n- Indirect Prompt Injection Surface: The skill retrieves and processes data from external web services, incorporating it into the agent's context.\n
- Evidence:\n
- Ingestion points: Data is fetched using the
requestslibrary inscripts/rest_request.py.\n - Boundary markers: The skill does not currently use specific delimiters to separate external data from system instructions in its output.\n
- Capability inventory: The script has the ability to write to the filesystem and make network requests.\n
- Sanitization: The
_compactfunction provides structural limiting but does not filter for potential instruction injection patterns.\n - Context: This is a standard consideration for skills that process third-party data, as external content could theoretically influence agent behavior.
Audit Metadata