The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest, score, and cluster records from external sources to resolve identities. This workflow presents an attack surface where malicious instructions could be embedded within identity fields (e.g., names, emails, or metadata) to influence the agent's behavior during resolution or merge decisions.
Ingestion points: The skill ingests records from "any source" as part of its core resolution mission.
Boundary markers: The skill includes instructions for "Tenant Isolation" and specifies that "PII is masked by default," which helps in limiting data leakage but does not fully prevent prompt-based manipulation of the scoring logic.
Capability inventory: The skill primarily performs logical processing; the included Python snippets focus on scoring and normalization (e.g., nickname expansion, phone normalization) and do not exhibit file system or network access capabilities.
Sanitization: The skill employs normalization techniques, such as regex-based character stripping and lowercase conversion, which provide a basic level of input processing.

identity-graph-operator