The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because its primary function is to scrape and process content from arbitrary, untrusted URLs provided as arguments. The instructions command the agent to extract verbatim phrases and analyze the site's positioning without defining boundary markers or sanitization procedures.
Ingestion points: Website content is ingested via website-scraper, WebFetch, and multiple Apify actors targeting the user-supplied agency URL.
Boundary markers: The skill lacks delimiters or instructions to ignore embedded commands within the scraped content.
Capability inventory: The skill writes analyzed data to Notion and prepares 'verbatim' signals for subsequent GTM analysis steps.
Sanitization: No sanitization or escaping of external content is specified before the data is processed or stored.
[CREDENTIALS_UNSAFE]: The skill instructions specify a workflow for utilizing a LinkedIn session cookie (li_at) retrieved from userConfig.linkedin_li_at. While this is used to facilitate deeper data scraping (e.g., headcount trends and employee rosters), handling live session tokens within an agent's operational logic increases the risk of credential exposure if the agent's context is compromised.
[EXTERNAL_DOWNLOADS]: The skill relies on a wide array of external third-party services and cloud-based scraping actors. Specifically, it references apify/website-content-crawler, scrapemint/linkedin-company-employees-scraper, harvestapi/linkedin-company-employees, and tri_angle/similarweb-scraper. The orchestration of multiple external execution environments for data processing expands the overall trust surface of the skill.

01-intake