firecrawl-search

Warn

Audited by Gen Agent Trust Hub on Nov 12, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill uses shell instructions to dynamically resolve the file paths of local scripts (firecrawl_search.py and research_wiki.py) by checking several locations, including repository roots and configuration files like ~/.aris/repo, before executing them with python3.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests full-page markdown content from arbitrary web URLs, creating a significant surface area for malicious instructions hidden in external data.
  • Ingestion points: Search results and scraped markdown content returned by the Firecrawl API.
  • Boundary markers: The skill includes a mandatory 'Evidence boundary' footer and explicit 'Key Rules' instructing the agent to treat all scraped content as untrusted data.
  • Capability inventory: The skill has access to the Bash tool and can write to the local research-wiki/ directory via a helper script.
  • Sanitization: Inputs such as arXiv IDs and URLs are validated using regular expressions (e.g., ^https?://[^[:space:]'"$\ ]+$) and wrapped in single quotes to mitigate simple injection attacks.
  • [COMMAND_EXECUTION]: The instructions direct the agent to construct and execute shell commands that incorporate user-provided search queries and URLs, relying on single-quoting and regex-based filtering to prevent command injection within the Bash tool.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 12, 2026, 06:07 AM