dashboard-forge

Pass

Audited by Gen Agent Trust Hub on Nov 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to automate and scrape vendor dashboards (e.g., GitHub, Azure, Cloudflare). This creates a vulnerability where malicious instructions embedded in web page content (such as application names, descriptions, or user-provided metadata) could be processed by the agent.\n
  • Ingestion points: The skill extracts content from the browser DOM using Playwright locators and page.evaluate, and generates screenshots (shot()) for agent analysis.\n
  • Boundary markers: The instructions do not specify any delimiters or explicit warnings to the agent to ignore instructions embedded in the dashboard content.\n
  • Capability inventory: The skill uses Bash, Read, Grep, and Glob tools, and describes executing local scripts and making network requests via curl.\n
  • Sanitization: While the skill emphasizes using a vault for secrets, it lacks specific sanitization or filtering for natural language instructions found in the web UI.\n- [DATA_EXFILTRATION]: The skill facilitates the capture of sensitive credentials (OAuth tokens, API keys, bearer tokens) and accesses local browser session data.\n
  • Credential Capture: It describes capturing bearer tokens "off the wire" (captureBearer) and extracting secrets from the DOM. Although it enforces a "secrets discipline" using a vault, the capability to move sensitive tokens is central to its operation.\n
  • Sensitive Path Access: The skill describes "seeding" profiles by copying session cookies from ~/Library/Application Support/Google/Chrome, which are credential-equivalent. It also references ~/.local/share/web-forge/ for storing live sessions.\n- [COMMAND_EXECUTION]: The skill automates workflows by executing local shell commands and scripts using Bun, Node.js, and osascript (AppleScript). This includes launching browsers with specific debugging flags and programmatically interacting with other browser windows.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 7, 2026, 02:12 AM