skills/sungjunlee/skills/gosu-review/Gen Agent Trust Hub

gosu-review

Pass

Audited by Gen Agent Trust Hub on Sep 29, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core function of processing untrusted external content from target files and web searches. \n
  • Ingestion points: The skill reads untrusted data from the user-supplied (which can include code files, diffs, and project artifacts), the Origin documentation (issues, plans, or requests), and external data fetched via web searches for the Alternatives vantage. \n
  • Boundary markers: The subagent prompts utilize plain text headers (e.g., Target, Origin, Context) to structure the brief but lack explicit delimiters or instructions to ignore potential commands or directives embedded within these dynamic fields. \n
  • Capability inventory: Subagents are configured with capabilities to "read whole files" and "search the web." The instructions explicitly suggest that subagents should "try it if you can" regarding the target, which increases the impact of a successful injection if the environment supports execution. \n
  • Sanitization: The skill does not perform any sanitization, filtering, or escaping of the ingested data before it is interpolated into the prompts for the independent subagents.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 29, 2026, 01:13 AM