solve-challenge
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is exposed to indirect prompt injection due to its core function of ingesting untrusted files and strings from CTF challenges while maintaining broad system capabilities.
- Ingestion points: External challenge bundles, files, remote server responses, and URLs parsed through
$ARGUMENTSor network commands inSKILL.md. - Boundary markers: Absent. There are no explicit boundaries or data isolation mechanisms configured to separate untrusted challenge text from agent instructions.
- Capability inventory: Broad tool capabilities enabling arbitrary command execution, file manipulation, and network interaction via the Bash tool (
file,strings,binwalk,checksec,curl,nc). - Sanitization: Absent. Content retrieved from external challenge components is passed directly to analysis utilities and into the agent context without filtration.
Audit Metadata