deepxiv
Warn
Audited by Gen Agent Trust Hub on May 18, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill takes user-supplied arguments from the
$ARGUMENTSvariable and inserts them directly into bash command strings inStep 3(e.g.,python3 tools/deepxiv_fetch.py search "QUERY"). This pattern is susceptible to command injection via shell metacharacters or command substitution (e.g.,$(...)) if the input is not strictly sanitized by the agent before execution. - [COMMAND_EXECUTION]: In
Step 6, the skill includes a complex bash script that dynamically resolves the file path forresearch_wiki.pyusingawkto parse local configuration files andgitto locate the repository root. This dynamic resolution and subsequent execution withpython3constitutes dynamic code loading from computed paths, which is a security risk if project files are tampered with. - [EXTERNAL_DOWNLOADS]: The documentation suggests that users install the
deepxiv-sdkpackage viapip. While this is required for the skill to function, it introduces a dependency on an external third-party package that should be verified for safety. - [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface.
- Ingestion points: Untrusted data from academic papers and metadata is retrieved via the
deepxivCLI and processed by the agent. - Boundary markers: While double quotes are used in some bash commands, there are no explicit instructions or delimiters used to ensure the agent ignores malicious instructions embedded within the retrieved paper content.
- Capability inventory: The skill is granted
Bash(*),Read, andWritecapabilities, and it performs several subprocess executions and file operations. - Sanitization: There is no evidence of sanitization or validation of the retrieved external content before it is displayed to the user or stored in the local wiki.
- [CREDENTIALS_UNSAFE]: The skill documentation notes that the
deepxivCLI automatically stores authentication tokens in the~/.envfile. Interacting with or modifying the.envfile is sensitive as it is a common location for high-value system secrets.
Audit Metadata