The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data (Lighthouse/PageSpeed reports) provided by the user in Phase 1. An attacker could theoretically embed malicious instructions within these reports to influence the agent's subsequent recommendations.
Ingestion points: The PHASE 1 — Ingest & Classify section describes extracting data from pasted text, screenshots, or URLs provided by the user.
Boundary markers: The skill body does not specify the use of delimiters (like XML tags or triple quotes) or explicit instructions for the agent to ignore directives found within the analyzed data.
Capability inventory: The agent has the capability to generate shell commands (e.g., for image conversion and icon resizing) and provide configuration changes for deployment files like netlify.toml and vercel.json based on the input.
Sanitization: There are no explicit instructions provided to sanitize or validate the content of the ingested reports.
[COMMAND_EXECUTION]: The skill provides several shell command snippets for common performance tasks.
Evidence: Instructions in Fix Batch 1 include commands for cwebp, sharp (via Node.js), sips (macOS built-in), and python3 (using the Pillow library). These are provided as code examples for the user to execute manually in their environment.
[EXTERNAL_DOWNLOADS]: The skill references external assets and services for performance optimization.
Evidence: Section F1.8 suggests downloading font files from Google Fonts (fonts.googleapis.com), which is a well-known and trusted service. Section F3.1 mentions the api.rss2json.com service as an example of an external API call that might be present in a user's critical path.

pagespeed-enhancer