diagnosing-bugs
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill encourages the agent to dynamically generate and execute scripts for testing and feedback loops, including bash scripts, curl invocations, and browser automation scripts (SKILL.md).
- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the processing of external debugging artifacts.
- Ingestion points: The agent is instructed to read GLOSSARY.md, ADRs, error logs, HAR files, and crash dumps provided by the user (SKILL.md).
- Boundary markers: The skill does not explicitly define delimiters for these external artifacts, though it instructs the agent to redact secrets.
- Capability inventory: The agent can execute shell commands, run curl scripts, drive headless browsers, and execute test harnesses (SKILL.md, scripts/hitl-loop.template.sh).
- Sanitization: No specific sanitization or escaping of external content is mentioned beyond the redaction of sensitive credentials.
Audit Metadata