The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's documentation in SKILL.md directs users to a CLI tool at webperf-snippets.net/CLI, which has been flagged as a phishing threat by automated security scanners.
[OBFUSCATION]: All 34 JavaScript files in the scripts/ directory, such as scripts/Back-Forward-Cache.js and scripts/FCP.js, are minified. This obscures the logic and prevents verification of the scripts' safety and behavior.
[REMOTE_CODE_EXECUTION]: The skill instructs users to run npx webperf-snippets, which downloads and executes external code from a remote registry at runtime, introducing supply chain risk.
[INDIRECT_PROMPT_INJECTION]: The skill audits arbitrary web pages by extracting data from the DOM and performance APIs. 1. Ingestion points: The skill uses document.querySelectorAll and performance.getEntriesByType across its scripts to ingest untrusted data. 2. Boundary markers: There are no explicit markers or instructions to prevent the agent from following malicious instructions embedded in audited pages. 3. Capability inventory: The skill has the ability to perform fetch requests and is executed via a powerful browser evaluation tool. 4. Sanitization: No sanitization or filtering is applied to the extracted web data before it is returned to the agent context.

webperf-loading