problem-dive

Pass

Audited by Gen Agent Trust Hub on Sep 6, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze untrusted external data, including system logs, trace/span chains, and user feedback reports, to perform root cause analysis. This creates a surface for indirect prompt injection if the analyzed logs or trace data contain malicious instructions intended to manipulate the agent's behavior during the investigation.
  • Ingestion points: System logs (e.g., tail <log>), trace/span chains, and user feedback strings defined in SKILL.md.
  • Boundary markers: The instructions do not specify the use of delimiters or provide warnings to ignore embedded instructions within the logs being read.
  • Capability inventory: The agent is expected to use tools for file reading, network searching for "due diligence" (尽调调研), and command execution for "experimental verification" (实测证伪).
  • Sanitization: There are no instructions for sanitizing or escaping content from external logs or records before processing.
  • [COMMAND_EXECUTION]: The methodology directs the agent to execute commands for evidence gathering (e.g., tail <log> | jq) and to perform "experimental verification" (实测证伪) to validate root cause hypotheses. This pattern involves executing shell commands or code based on data derived from the analysis process.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 6, 2026, 09:28 PM
Security Audit — agent-trust-hub — problem-dive