The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill provides explicit instructions to never hardcode API credentials, tokens, or tenant identifiers. It mandates the use of environment-bound authentication and rotation policies.
[DATA_EXFILTRATION]: Instructions strictly forbid logging clinical content or credentials. The skill limits network operations via WebFetch to official documentation sources only.
[EXTERNAL_DOWNLOADS]: The skill references documentation from Abridge's official domains and the U.S. Department of Health and Human Services, which are well-known and appropriate for the skill's stated purpose.
[INDIRECT_PROMPT_INJECTION]: The skill addresses potential injection risks from external specifications.
Ingestion points: Repository configuration, adapters, tests, and integration specifications (SKILL.md).
Boundary markers: Mandates typed request/response envelopes, strict parsing, and explicit unknown-field behavior.
Capability inventory: Uses Read, Glob, Grep, WebFetch, Write, and Edit tools (SKILL.md).
Sanitization: Explicitly requires redaction of clinical content, credentials, and tenant identifiers from logs and fixtures.

abridge-sdk-patterns