The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from both CRM records and public web research, creating an attack surface where malicious content in those sources could potentially influence agent behavior.
Ingestion points: Public company research results (hiring, funding, leadership signals) in Step 7 and CRM activity/opportunity records in Step 5.
Boundary markers: The instructions provide specific logic for judging engagement and drafting emails based on canonical use cases, which acts as a functional boundary, though no explicit sanitization or "ignore embedded instructions" delimiters are used for external content.
Capability inventory: The agent has the capability to execute tool routines (enrich-company, cpj-find-lists-of-people), query CRM databases, and perform local file-write operations for configuration storage (Step 12).
Sanitization: The skill does not explicitly describe mechanisms to sanitize or escape external data before it is interpolated into the prompts used to generate the email sequence.
[COMMAND_EXECUTION]: The skill instructions require the agent to verify the execution environment and perform tasks using the clay CLI and connected tool routines.
The agent uses clay whoami and clay routines get to check status and pricing.
It invokes specific routines like enrich-company and cpj-find-lists-of-people for data retrieval.
These are identified as vendor-owned resources for 'clay-run' and are consistent with the skill's primary function.

prospect-one-account