creating-debug-tests-and-iterating

Pass

Audited by Gen Agent Trust Hub on Apr 30, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill uses pseudo-system tags like <system-reminder> and markers like *CRITICAL* to instruct the agent to override its standard behavior, specifically demanding it ignore existing tests and disregarded prior constraints.
  • [COMMAND_EXECUTION]: The skill explicitly instructs the agent to create and run shell scripts and background processes (e.g., npm run dev, python server.py) using subprocess.run and exec, which facilitates arbitrary command execution on the host system.
  • [INDIRECT_PROMPT_INJECTION]: The debugging loop requires the agent to read and analyze logs and script outputs from the application under test. This creates a vulnerability surface where a malicious application could output instructions that influence the agent's next actions.
  • Ingestion points: Application logs and the standard output/error of generated debug scripts (SKILL.md).
  • Boundary markers: None. The agent is not instructed to treat log content as data rather than instructions.
  • Capability inventory: Subprocess execution, shell command execution, and file writing across Bash, Python, and Node.js examples (SKILL.md).
  • Sanitization: None. The instructions tell the agent to "Analyze the output: read logs, identify errors, do whatever you need to."
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 30, 2026, 10:13 PM
Security Audit — agent-trust-hub — creating-debug-tests-and-iterating