drupal-manual-video-testing

Warn

Audited by Gen Agent Trust Hub on Jul 27, 2026

Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONREMOTE_CODE_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because its core workflow involves fetching and following instructions from external drupal.org issues and merge requests, which are attacker-controllable sources.
  • Ingestion points: The skill reads external instructions from drupal.org/project/<module>/issues/<id> or drupal.org/node/<id> in Chapter 2 of SKILL.md.
  • Boundary markers: No explicit delimiters or security markers are used to isolate or distinguish external content from trusted system instructions.
  • Capability inventory: The skill possesses extensive capabilities, including shell command execution via ddev exec (in lib.sh), environment modification in preflight.sh, and full browser automation via agent-browser (in session.sh).
  • Sanitization: There is no evidence of filtering or sanitization of the external content before it is interpreted by the agent to drive browser actions or execute commands.
  • [DATA_EXFILTRATION]: The browser session configuration in session.sh includes flags (--remote-debugging-address=0.0.0.0 and --remote-allow-origins=*) that expose the Chromium DevTools Protocol port to all network interfaces within the container environment, which could be exploited by other processes to gain control of the browser session.
  • [REMOTE_CODE_EXECUTION]: The skill uses agent-browser to execute dynamic JavaScript code (eval) within the Chromium session as described in SKILL.md. This dynamic execution is driven by the agent's interpretation of external testing instructions, presenting a pathway for executing arbitrary code in the browser context.
  • [COMMAND_EXECUTION]: The skill makes extensive use of ddev exec to run commands inside the container. It also establishes persistence within the DDEV environment by creating a configuration file (.ddev/config.manual-testing.yaml) that includes automated hooks to re-install dependencies on every container start.
  • [EXTERNAL_DOWNLOADS]: The preflight.sh script performs downloads from trusted and well-known sources, including the official Google Fonts repository for the Montserrat font and the npm registry for the agent-browser package.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 27, 2026, 01:24 PM
Security Audit — agent-trust-hub — drupal-manual-video-testing