The Agent Skills Directory

[Command Execution] (SAFE): The bash scripts execute lighthouse, curl, and python3 for their intended purposes. Commands are constructed using variables within double quotes to mitigate simple shell injection.
[Unverifiable Dependencies & Remote Code Execution] (SAFE): The script scripts/lighthouse.sh recommends installing the lighthouse package via the official NPM registry, which is a trusted source.
[Data Exposure & Exfiltration] (SAFE): Data is sent via curl to googleapis.com, which is a trusted domain. The PageSpeed API key is required as an environment variable (PAGESPEED_API_KEY) rather than being hardcoded in the source, preventing accidental credential exposure.
[Indirect Prompt Injection] (LOW): The skill fetches data from arbitrary external URLs provided by the user or agent. If an agent subsequently reads the generated JSON reports, it could ingest untrusted content embedded in those pages.
Ingestion points: Target URLs processed in scripts/lighthouse.sh and scripts/pagespeed.sh.
Boundary markers: None; results are saved as raw JSON reports.
Capability inventory: curl (network), lighthouse (CLI execution).
Sanitization: Filenames are sanitized using sed to remove protocol and special characters; URLs are encoded using a Python helper.
[Dynamic Execution] (LOW): scripts/pagespeed.sh uses an inline Python heredoc to perform URL encoding. While this involves generating script content at runtime, it uses a static template and is considered low risk.

seo-aeo-audit