guarded-websearch-codex
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONOBFUSCATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses a Bash script
quarantine-search-codex.shto execute thecodexCLI with search capabilities. The execution is performed within a restricted environment using--sandbox read-onlyand--ephemeralflags to prevent persistent changes or unauthorized file system access. The script includes input validation to block shell metacharacters like backticks or subshell expansion in the search query. - [OBFUSCATION]: The
sanitize.tsscript includes logic to detect and remove zero-width characters, bidirectional overrides, and other invisible Unicode characters from search results. This is a security feature designed to prevent 'invisible character' prompt injection attacks where malicious instructions are hidden in plain sight. - [INDIRECT_PROMPT_INJECTION]: The skill is designed specifically to mitigate indirect prompt injection vulnerabilities. It ingests untrusted data from web search results (titles and snippets) and passes them through a static sanitizer (
sanitize.ts) that filters for known attack patterns like instruction overrides ('ignore previous instructions') and chat template markers. It provides mandatory boundary reporting viaaggregate_flagsso the main agent can assess the safety of the results.
Audit Metadata