The Agent Skills Directory

[Prompt Injection] (LOW): The skill possesses an indirect prompt injection surface through the use of the Claygent AI research agent.
Ingestion points: Untrusted external data enters the context via variables such as {{company_url}}, {{company_domain}}, and {{LinkedIn URL}} in claygent/SKILL.md and company-enrichment/SKILL.md.
Boundary markers: Instructions in claygent-guide.md provide structural output requirements (e.g., 'Return format: [text/number/URL/true-false]') but do not include security-specific delimiters or instructions to ignore embedded commands within the scraped content.
Capability inventory: The skill facilitates web browsing and scraping (via Claygent) and JavaScript execution (via Clayscript) across its sub-skills.
Sanitization: While basic data cleaning formulas are provided in copy-paste-formulas.md, there is no evidence of sanitization or escaping of external web content before it is processed by the AI models.
[Dynamic Execution] (SAFE): The skill extensively uses Clayscript, which is a JavaScript-based formula language for data manipulation in Clay. As documented in clayscript-guide.md and conditional-logic/SKILL.md, this execution is limited to the platform's sandbox for intended data transformation purposes (e.g., cleaning names, calculating scores) and does not exhibit malicious patterns.
[Data Exposure] (SAFE): While 'Clay API keys' are mentioned in clay-operations/SKILL.md and clay-operations-credit-optimization.md, these references are conceptual instructions for users to connect their own accounts. No hardcoded secrets or sensitive local file paths were found.

clay