The Agent Skills Directory

[SAFE]: The skill consists of instructional Markdown files that adhere to security best practices. It includes a comprehensive skill contract that enforces user authorization for any side effects and emphasizes evidence-based reasoning.
[NO_CODE]: This skill contains no executable scripts, binaries, or automated code installation procedures, significantly reducing its attack surface.
[INDIRECT_PROMPT_INJECTION]: The skill architecture explicitly addresses the risk of indirect prompt injection from untrusted data sources.
Ingestion points: The skill is designed to ingest data from user context, project files, and external sources such as web scrapes and platform exports.
Boundary markers: The instructions in SKILL.md provide a clear boundary by stating that scraped text and third-party exports must be treated as untrusted input.
Capability inventory: The skill uses tools to read marketing platform data and write recommendations, but requires authorization for any state-changing actions.
Sanitization: The agent is instructed to ignore any instructions embedded within the evidence it processes, effectively preventing the execution of malicious commands hidden in external data.

ad-account-auditor