investigate-bug-report

Pass

Audited by Gen Agent Trust Hub on Jul 29, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted bug reports and instructions to reproduce defects, creating a surface for indirect prompt injection where malicious instructions could be embedded in the report data.
  • Ingestion points: User-provided bug reports, observed behaviors, and reproduction steps defined in Step 1 of SKILL.md.
  • Boundary markers: Absent; there are no instructions or delimiters used to prevent the agent from executing commands embedded within the external report data.
  • Capability inventory: The skill is configured with Read, Glob, Grep, and Bash tools, allowing for substantial file system access and shell command execution.
  • Sanitization: Absent; the skill directs the agent to run user-provided failing commands or minimal scripts in Step 2 without any validation or sanitization mechanisms.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 29, 2026, 12:04 PM
Security Audit — agent-trust-hub — investigate-bug-report