web-perf
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external web pages, creating a potential surface for indirect prompt injection attacks.\n
- Ingestion points: Data enters the agent's context through
navigate_pageandtake_snapshotcalls during Phase 1 and Phase 4 of the audit workflow.\n - Boundary markers: Absent. The instructions do not define clear delimiters or provide the agent with rules to distinguish between its internal instructions and potentially adversarial content embedded in audited pages.\n
- Capability inventory: The agent possesses capabilities including network navigation, performance tracing, and the ability to read local project configuration files (e.g.,
package.json,webpack.config.js).\n - Sanitization: Absent. No explicit methods for escaping, validating, or filtering the ingested web content are described in the skill instructions.\n- [EXTERNAL_DOWNLOADS]: The skill references external documentation and tooling resources.\n
- It retrieves current performance guidelines and metric thresholds from official documentation sources, including
web.devanddeveloper.chrome.com.\n - It recommends the installation of the
chrome-devtools-mcppackage vianpxto enable the required tool functionality.
Audit Metadata