gs-navigate-pages
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill extracts search result data (titles, authors, and snippets) from Google Scholar, which is attacker-controlled external content that could contain malicious instructions.
- Ingestion points: The
evaluate_scriptlogic inSKILL.md(Step 3) extracts text content from the#gs_res_cclcontainer. - Boundary markers: Absent. The reporting template in Step 4 does not utilize clear delimiters or instructions to the agent to treat the search result content strictly as data.
- Capability inventory: The skill utilizes
mcp__chrome-devtools__navigate_pageandevaluate_script, allowing for further browser navigation and script execution based on agent decisions. - Sanitization: Extracted snippets are truncated to 200 characters and trimmed, but no validation or filtering for potential injection patterns is performed.
Audit Metadata