The Agent Skills Directory

[SAFE]: The skill defines strict data isolation and confidentiality boundaries, instructing the agent to omit credentials, private relationship details, and sensitive financial data from reusable outputs.
[PROMPT_INJECTION]: The skill processes external information regarding investor profiles and meeting feedback, creating a surface for indirect prompt injection. This is mitigated by the skill's internal controls and lack of dangerous tools.
Ingestion points: Untrusted data enters via investor fit records and meeting notes (references/targeting-meetings.md), and update contracts (references/updates-cadence.md).
Boundary markers: Instructions require tagging data with permitted_audience and confidentiality_basis.
Capability inventory: The skill lacks active tools or script execution capabilities; its primary function is generating structured drafts and reports.
Sanitization: The workflow requires immutable versioning, legal review, and explicit authorization gates before any data is sent.

investor-relations