anthropic-webapp-testing

Warn

Audited by Gen Agent Trust Hub on May 16, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/with_server.py uses subprocess.Popen with shell=True to execute command-line strings provided as arguments. It also uses subprocess.run to execute trailing commands. This allows for arbitrary shell command execution within the agent's environment.
  • [PROMPT_INJECTION]: The skill instructions in SKILL.md advise the agent to avoid reading the source code of utility scripts ('DO NOT read the source until you try running the script first'), which functions as a concealment pattern. The skill also presents an indirect prompt injection surface through its primary function of processing external web content.
  • Ingestion points: Data from external web applications is ingested via Playwright using page.content(), page.screenshot(), and element discovery tools.
  • Boundary markers: The skill does not implement delimiters or instructions to isolate untrusted web data from the agent's instruction context.
  • Capability inventory: The agent is equipped with shell command execution tools (scripts/with_server.py) and file system write access for logs and screenshots.
  • Sanitization: No content sanitization or instruction-filtering is applied to the data retrieved from web pages before it is processed by the agent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 16, 2026, 07:15 PM
Security Audit — agent-trust-hub — anthropic-webapp-testing