sst-llm-judge-ranker

Pass

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and processes the contents of untrusted files within an LLM comparison prompt.
  • Ingestion points: The skill ingests file content via the 'Read' tool from paths or directories provided in the 'Direct compare', 'Batch rank', and 'Incremental insert' modes.
  • Boundary markers: The judge prompt uses simple delimiters like 'Artifact A:' and 'Artifact B:' without specific instructions to the agent to treat the content as inert data or to ignore embedded commands.
  • Capability inventory: The skill performs local file reads and writes ranking results to JSON and JSONL files on the local disk. It does not utilize network access or shell execution.
  • Sanitization: No sanitization, escaping, or validation of the artifact content is performed before it is interpolated into the comparison template.
  • [NO_CODE]: This skill consists entirely of markdown-based logic and prompt templates, containing no executable scripts, binaries, or external code dependencies.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 16, 2026, 01:00 AM
Security Audit — agent-trust-hub — sst-llm-judge-ranker