The Agent Skills Directory

[COMMAND_EXECUTION]: The skill launches a headless Chrome instance using the chrome-launcher package. It defaults to using the --no-sandbox flag, which is often required in containerized or CI environments but reduces the security isolation of the browser process when handling untrusted web content.
[EXTERNAL_DOWNLOADS]: The skill's primary function is to fetch and audit external URLs. This involves crossing a network boundary to download and render content from potentially untrusted third-party servers.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It retrieves information from audited websites (such as audit titles and element snippets) and incorporates this data into a Markdown report designed for processing by AI agents. A malicious website could embed instructions in its content that, once captured by the audit, might influence the behavior of the agent reading the report.
Ingestion points: Data from external sites is ingested via the lighthouse library in src/api.ts and src/lighthouse-runner.ts.
Boundary markers: The report generated in src/report.ts lacks explicit delimiters or warnings to help the consuming agent distinguish between technical audit metadata and content scraped from the target site.
Capability inventory: The skill allows agents to execute the lightscore CLI and write report files to the local filesystem.
Sanitization: There is no evidence of sanitization or escaping of the scraped content before it is placed into the Markdown report.

auditing-lighthouse-scores