The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data including company domains, LinkedIn URLs, and headcount counts.
Ingestion points: Input variables company_domain and company_linkedin_url (in SKILL.md and clay-workflow.md).
Boundary markers: The LLM prompt in §6 utilizes a structured JSON response format and explicit delimiters for input data.
Capability inventory: The skill uses gpt-4o-mini for text generation and various Clay table actions for data retrieval.
Sanitization: High. Deterministic formulas in clay-table.md (e.g., Hiring Surge Dept) map tool outputs to a strict set of predefined string constants ("sales", "marketing", or empty) before they are passed into the LLM prompt, effectively neutralizing any injection attempts from external data sources.
[COMMAND_EXECUTION]: The documentation includes standard CLI commands for the Clay platform (clay workflows, clay routines). These are instructional examples for users to deploy and manage the skill within the supported environment and do not represent malicious execution.
[DATA_EXPOSURE]: The skill performs lookups on company identifiers for employee headcount enrichment. This is a core feature of the automation and is performed using standard platform tools with clear cost and privacy guidance for the operator.
[DYNAMIC_EXECUTION]: clay-workflow.md contains Python code snippets intended for execution within the Clay workflow environment. The code implements logical operations such as string normalization, growth percentage calculations, and output quality control (QC). The logic is transparent, localized, and contains no unsafe dynamic loading or external script execution.

playbook-hiring-surge