The Agent Skills Directory

[SAFE]: The skill implements strong safety boundaries by explicitly stating that tool invocation does not authorize network access, credential retrieval, or data transfer. It requires data owner approval before exporting telemetry or implementing caches.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it ingests untrusted data from local documentation, configuration, and fixtures using Read, Glob, and Grep tools, and has the capability to write to the repository using Write and Edit.
Ingestion points: Reads documentation, configuration files, and code fixtures (SKILL.md).
Boundary markers: The instructions explicitly demand the use of "content-free telemetry" and the removal of secrets and real notebook content from fixtures.
Capability inventory: Uses Write and Edit for repository artifacts; no network or shell execution capabilities are enabled in allowed-tools.
Sanitization: Relies on manual redaction of request evidence and aggregated data processing to prevent sensitive data leakage.

onenote-cost-tuning