diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which constitutes a potential injection surface. 1. Ingestion points: Reads from GLOSSARY.md, ADRs, test outputs, curl responses, logs, HAR files, and trace replays. 2. Boundary markers: None mentioned for external data. 3. Capability inventory: Executes shell commands, runs tests, and creates scripts. 4. Sanitization: Relies on agent-side redaction of secrets to
<REDACTED>. - [COMMAND_EXECUTION]: The agent is instructed to run various shell-based diagnostic tools, including test suites, curl, and a human-in-the-loop bash script template (
scripts/hitl-loop.template.sh) to facilitate debugging. - [DYNAMIC_EXECUTION]: The methodology encourages the creation and execution of dynamic code such as bisection harnesses, throwaway harnesses, and temporary instrumentation logs to reproduce and isolate software bugs.
Audit Metadata