sn-deepresearch-cli
Warn
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions provide shell command templates that interpolate raw user input (
<query>) directly into command arguments (e.g.,deepresearch "<query>"). This pattern is susceptible to shell command injection if the input contains characters like backticks or subshell markers, especially as the instructions emphasize retaining the original user text. - [PRIVILEGE_ESCALATION]: The skill requires the agent to grant extensive permissions, including 'read', 'write', 'exec', and 'process' capabilities to its working environment. It also utilizes global package installation (
npm install --global), which introduces binaries to the system path and typically requires elevated permissions. - [EXTERNAL_DOWNLOADS]: The skill executes global installations of the
sensenova-skills-deepresearchand@agentclientprotocol/claude-agent-acppackages. While the main package is a vendor-owned resource, this process fetches and executes code from a remote registry. - [INDIRECT_PROMPT_INJECTION]: The skill handles untrusted data from user queries and external web pages via the Camofox tool. Evidence Chain: 1. Ingestion points: User queries and scraped web content. 2. Boundary markers: Double quotes are used in shell templates, but no specific sanitization or delimiters are applied to the content. 3. Capability inventory: The skill has broad capabilities including shell execution, file system writes, and network operations. 4. Sanitization: No sanitization or filtering logic is specified in the agent instructions.
Audit Metadata