diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from network traces, event logs, and user-provided terminal inputs through the hitl-loop.template.sh script.
- Ingestion points: Network requests, payloads, and terminal inputs captured by the read command in the human-in-the-loop script.
- Boundary markers: The agent is instructed to quote only the lines that carry the signal and redact every secret first.
- Capability inventory: The skill possesses the ability to execute shell commands, curl requests, and run dynamically generated test scripts.
- Sanitization: While secret redaction is explicitly required, there is no specific sanitization described for filtering potential prompt injection attempts within the captured data.
- [DYNAMIC_EXECUTION]: The workflow involves the dynamic creation and execution of scripts, including Bash reproduction harnesses and browser-based tests, which are used to confirm and minimize bug reports at runtime.
- [COMMAND_EXECUTION]: The skill makes extensive use of CLI tools and utilities, such as curl, git, and custom shell scripts, to interact with the development environment and verify application behavior.
Audit Metadata