skills/maxgfr/leboncoin-cdp/leboncoin/Gen Agent Trust Hub

leboncoin

Warn

Audited by Gen Agent Trust Hub on Oct 8, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DATA_EXPOSURE]: The skill automatically copies the user's primary browser profile (containing cookies, session tokens, history, and preferences) from standard system paths to a local hidden directory (~/.lbc-scraper/profile). This practice creates a local clone of sensitive authentication data.
  • [COMMAND_EXECUTION]: The script executes shell commands to manage the browser environment:
  • Spawns browser binaries (Chrome, Brave, or Chromium) with debugging flags enabled (--remote-debugging-port).
  • Uses pgrep via execSync to detect and manage running browser processes.
  • [DYNAMIC_EXECUTION]: The core automation engine relies on Runtime.evaluate to execute dynamically generated JavaScript strings within the context of the browser tab. This allows the skill to bypass standard automation detection by mimicking in-page script behavior.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external classified listings which presents a potential injection surface.
  • Ingestion points: Listing descriptions and titles are scraped from the target website using Next.js data routes (see src/scraper.ts).
  • Boundary markers: External data is structured into comparables.json and rendered into comparables.md using markdown table and code block delimiters.
  • Capability inventory: The skill can write files to the local system, launch processes, and perform browser-based network operations.
  • Sanitization: The skill extracts structured data from JSON/HTML but does not perform content filtering for adversarial instructions within listing descriptions.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 8, 2026, 09:47 PM