diagnosing-bugs

Pass

Audited by Gen Agent Trust Hub on Sep 22, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest and analyze potentially untrusted external artifacts to build reproduction loops.
  • Ingestion points: SKILL.md describes a process of reading and replaying captured network traces, payloads, event logs, HAR files, and log dumps provided by users or external environments.
  • Boundary markers: The instructions do not specify the use of delimiters or explicit warnings to ignore instructions embedded within these external data artifacts.
  • Capability inventory: The agent is guided to execute shell scripts (scripts/hitl-loop.template.sh), CLI invocations, and network operations (curl) based on findings derived from analyzing these artifacts.
  • Sanitization: While the skill strongly mandates redacting sensitive credentials for privacy, it lacks specific sanitization measures to prevent embedded instructions from influencing the agent's subsequent logic or tool usage.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 22, 2026, 11:39 PM
Security Audit — agent-trust-hub — diagnosing-bugs