The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill processes untrusted data from external sources to identify sales opportunities, which presents a surface for potential injection attacks.
Ingestion points: The skill explicitly ingests data from email, chat, transcripts, and external docs, which are identified as untrusted content.
Capability inventory: The skill possesses the ability to create records (opportunities) within a CRM and communicate through email or chat connectors.
Sanitization & Boundary Markers: The instructions include robust mitigations, such as treating external content as data only (not instructions), reporting instruction-like text without acting on it, and suppressing links found within untrusted content to prevent unauthorized redirection.
Human-in-the-loop: The skill requires that all "content-originated" actions be shown to the user for approval before execution, even in scheduled runs.
[Agent Constraint Modification]: The instructions contain directives aimed at limiting the agent's internal refusal mechanisms in favor of the underlying connector's permission settings.
Evidence: The skill states: "never add a restriction the connector does not impose, and never refuse an action the user asked for on the plugin's own authority."
Context: While likely intended to ensure that the agent does not block legitimate tool usage that a specific CRM or Email connector allows, this instruction explicitly asks the agent to override its own safety or operational constraints if a user requests an action.

expansion-whitespace