The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill reads sensitive local credential files to authenticate with Google APIs.
Evidence: The script scripts/gsc-inspect.mts is designed to access ~/.config/gcloud/application_default_credentials.json and the file path specified in the GOOGLE_APPLICATION_CREDENTIALS environment variable to obtain access tokens for Google Search Console.
[EXTERNAL_DOWNLOADS]: The skill dynamically downloads a well-known external tool at runtime if it is missing from the environment.
Evidence: The script scripts/lighthouse-run.mts uses npx -y lighthouse to fetch and run the Google Lighthouse CLI from the npm registry as a fallback when a local binary is not found.
[COMMAND_EXECUTION]: The skill executes external CLI tools via subprocesses to gather SEO and performance data.
Evidence: Various scripts in the scripts/ directory invoke gcloud, lighthouse, and squirrel using Node.js execFileSync and execFile or shell execution.
[PROMPT_INJECTION]: The skill processes data from external websites, which provides a surface for indirect prompt injection attacks.
Ingestion points: External website content is ingested via crawler tools (Lighthouse and Squirrel) to generate technical audit reports.
Boundary markers: The instructions do not define clear boundary markers or instructions to ignore embedded content when processing data from these external crawls.
Capability inventory: The skill possesses significant capabilities, including the ability to execute shell commands, read/write files in the repository, and perform network requests to multiple APIs.
Sanitization: No specific sanitization logic is implemented to filter potentially malicious instructions from the ingested site content before it is processed by the agent.

seo