The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from external landing pages.
Ingestion points: The skill uses WebFetch to retrieve the full HTML content and visible text (headlines, subheadlines, trust signals) from landing page URLs provided by the user or extracted from Google Ads campaigns (SKILL.md, Phase 2).
Boundary markers: There are no explicit instructions or delimiters used to wrap the fetched HTML content to prevent the agent from following embedded malicious instructions.
Capability inventory: The skill possesses significant capabilities, including executing GAQL queries against a Google Ads account via runScript and writing audit data to the local file system in {data_dir}/landing-page-history.json.
Sanitization: The instructions do not define any sanitization, filtering, or escaping for the fetched HTML before it is analyzed for subjective qualities such as 'tone' or 'trust signals'.
[EXTERNAL_DOWNLOADS]: The skill fetches performance and accessibility metrics from a well-known service.
Evidence: The skill calls the official Google PageSpeed Insights API at https://www.googleapis.com/pagespeedonline/v5/runPagespeed via WebFetch (SKILL.md, Phase 2).
[COMMAND_EXECUTION]: The skill uses platform-specific tools to interact with Google Ads data.
Evidence: The skill utilizes runScript to execute Google Ads Query Language (GAQL) queries against ad_group_ad, ad_group, and keyword_view to resolve target pages and pull conversion metrics (SKILL.md, Phases 1 and 2).

google-ads-landing