semantic-scholar
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill executes bash commands that incorporate user-provided
$ARGUMENTS. While the instructions describe a parsing step, the raw execution in Step 2 (python3 "$S2_FETCHER" search "QUERY") poses a risk of command injection if the agent does not strictly sanitize the input before shell interpolation. Additionally, the skill resolves script paths based on environment variables and local files (.aris/installed-skills-codex.txt), which could be manipulated to execute unintended scripts. - [INDIRECT_PROMPT_INJECTION]:
- Ingestion points: Fetches paper titles, authors, abstracts, and TLDRs from the Semantic Scholar API (
api.semanticscholar.org). - Boundary markers: No explicit delimiters or safety instructions are used to distinguish retrieved academic content from system instructions.
- Capability inventory: The skill possesses
Bash,Read, andWritecapabilities, enabling it to write persistent files to theresearch-wiki/directory. - Sanitization: The skill does not define any sanitization or filtering of the retrieved data before it is presented to the user or stored in the wiki, creating a surface for adversarial instructions hidden in academic abstracts.
- [DYNAMIC_EXECUTION]: The skill provides a fallback mechanism that uses inline Python code with
urllibto interact with the API if the primary tool is missing. This runtime code generation and execution increases the complexity of the security boundary. - [EXTERNAL_DOWNLOADS]: Fetches academic literature data from Semantic Scholar's official API (
api.semanticscholar.org). This is a well-known service for research, and the data retrieval is essential to the skill's primary function.
Audit Metadata