scrape-to-crm

Pass

Audited by Gen Agent Trust Hub on Mar 6, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes data from external LinkedIn profiles and company pages, which are attacker-controllable sources.
  • Ingestion points: Scraped data is retrieved from various LinkedIn actors via the Apify MCP tools as described in SKILL.md.
  • Boundary markers: The skill does not define clear delimiters or use 'ignore embedded instructions' prompts when handling the scraped content.
  • Capability inventory: The skill utilizes curl commands to interact with the Attio API and can potentially build n8n workflows.
  • Sanitization: While the documentation suggests using JSON.stringify() to prevent breakage, the provided code examples use direct string interpolation (e.g., <full_name>) within shell commands, which may allow malicious profile content to influence agent logic.
  • [COMMAND_EXECUTION]: The skill executes curl commands to send data to the Attio CRM. While targeting a well-known service (api.attio.com), the inclusion of untrusted data from LinkedIn scrapes into these shell commands creates a risk of command injection if the agent does not properly escape the variables before execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 6, 2026, 05:27 PM