bug-reproduction-validator

Pass

Audited by Gen Agent Trust Hub on Mar 30, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it systematically processes untrusted data in the form of bug reports. A malicious actor could embed instructions within a report to influence the agent's actions during reproduction.
  • Ingestion points: Bug reports and issue descriptions enter the agent context (SKILL.md).
  • Boundary markers: Absent. The prompt lacks explicit delimiters to segregate user-provided data from system instructions.
  • Capability inventory: The agent can perform file exploration, examine logs and databases, and interact with the UI via the agent-browser CLI.
  • Sanitization: Absent. There is no mention of filtering or validating the content of the reports before investigation.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 30, 2026, 01:20 AM