The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides tools for Natural Language Processing (NLP) and PageSpeed analysis that ingest content from external, untrusted URLs, creating a surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context via the output of scripts/nlp_analyze.py, scripts/pagespeed_check.py, and scripts/gsc_inspect.py when processing external websites.
Boundary markers: The skill instructions do not establish explicit delimiters or include warnings for the agent to ignore instructions embedded within the analyzed content.
Capability inventory: The agent has access to sensitive capabilities including Google Search Console (read/write), Google Analytics 4 reporting, and the Google Indexing API, which could be targeted by adversarial content in analyzed pages.
Sanitization: No explicit sanitization, filtering, or validation of the remote content is described in the prompt instructions before the data is processed by the agent.
[COMMAND_EXECUTION]: The skill relies on the execution of several local Python scripts to perform its functions, which involves piping data from the agent to shell commands.
Evidence: The skill executes scripts such as google_auth.py, gsc_query.py, indexing_notify.py, and ga4_report.py to interact with Google's Cloud infrastructure and APIs.

seo-google