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) usingsubprocess.runandexec, 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