The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the user's workspace using the clay CLI. The Build.clay method in scripts/cascade_lib.py uses subprocess.run to execute commands such as clay workflows, clay functions, and clay triggers for graph management and deployment.
[DYNAMIC_EXECUTION]: The script scripts/test_codegen.py employs exec() and compile() to validate the Python code nodes generated for Clay workflows. This allows developers to test the enrichment logic locally against a mock context before committing changes to the live platform.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted contact data from webhooks or Clay tables, creating a potential vector for indirect prompt injection attacks where malicious instructions could be embedded in the enriched data fields.
Ingestion points: Data enters the workflow via the 0a Intake node, which accepts a flat payload object or direct inputs (scripts/build_cascade.py).
Boundary markers: The skill implements a strict internal record contract and source attribution system to maintain data integrity during the enrichment process (SKILL.md, references/cascade-config.md).
Capability inventory: The generated workflows have the capability to perform arbitrary HTTP requests via Clay's generic HTTP action and execute custom Python logic in code nodes.
Sanitization: The skill instructions acknowledge that the target execution environment is limited to basic string and JSON operations, which may restrict the implementation of complex sanitization filters.
[DATA_EXFILTRATION]: A core feature of the cascade is the callback_url, which allows the finished record—containing potentially sensitive PII like mobile numbers and work emails—to be POSTed to an external URL provided by the caller at runtime (scripts/build_cascade.py).

contact-unlimited-enrichment-cascade