The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves and displays search performance data that includes user-controlled strings such as search queries, which presents a surface for indirect prompt injection.
Ingestion points: The GoogleSearchConsoleClient class in client.ts fetches metrics and queries from the Google Search Console API in methods querySearchAnalytics and inspectUrl.
Boundary markers: Absent. API results are transformed into Markdown tables in formatters.ts without explicit delimiters or system instructions to ignore embedded commands.
Capability inventory: The skill is configured in SKILL.md to allow Bash, Read, Write, and Edit tools. The application code in client.ts performs network requests and config.ts performs local file reads.
Sanitization: Absent. There is no specific validation or filtering performed on the search queries or page metadata to detect or strip potential injection payloads.
[DATA_EXFILTRATION]: The skill performs expected network operations and accesses configuration files as part of its documented functionality.
It communicates with official Google domains including searchconsole.googleapis.com and oauth2.googleapis.com for data retrieval and authentication.
It reads local service account key files via fs.readFileSync in config.ts based on paths specified in environment variables such as GOOGLE_APPLICATION_CREDENTIALS.

google-search-console