parallel-web-search

Pass

Audited by Gen Agent Trust Hub on Mar 1, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes the parallel-cli tool via Bash, interpolating $ARGUMENTS and $FILENAME into the command string. While the agent is instructed to format the filename safely, the reliance on LLM-based sanitization for shell arguments poses a potential risk for command injection if input is not properly handled by the agent.\n- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by ingesting untrusted content from the web.\n
  • Ingestion points: Web search result excerpts and titles are read from JSON files in the /tmp/ directory.\n
  • Boundary markers: Absent. The skill does not utilize delimiters or provide explicit instructions to the agent to treat search result content as potentially malicious or to ignore embedded commands.\n
  • Capability inventory: The skill possesses capabilities to execute shell commands via parallel-cli.\n
  • Sanitization: None. The agent is directed to synthesize a response using extracted snippets without any filtering or escaping mechanisms.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 1, 2026, 12:37 AM