The Agent Skills Directory

[SAFE]: The skill defines an autonomous governor for self-improving software with a primary focus on cost control, financial stability, and system performance monitoring.
[DATA_EXFILTRATION]: No evidence of sensitive data access or unauthorized network exfiltration was found. The network operations mentioned are limited to the intended functionality of routing requests to LLM providers and scraping APIs.
[PROMPT_INJECTION]: The skill is designed to process external user data for shadow testing and model evaluation. While this constitutes an indirect injection surface, the risk is mitigated by instructions to use objective mathematical criteria for grading and strict asynchronous shadow testing environments.
Ingestion points: Real user data is processed as shadow traffic for model evaluation (SKILL.md).
Boundary markers: Explicit mathematical criteria (e.g., points for JSON formatting) are required before testing (SKILL.md).
Capability inventory: Autonomously updates routing weights, trips circuit breakers, and generates evaluation prompts (SKILL.md).
Sanitization: Employs 'LLM-as-a-Judge' grading to validate outputs against established production baselines (SKILL.md).
[REMOTE_CODE_EXECUTION]: No remote script downloads, piped execution, or dynamic code execution patterns were identified. The provided code example is a static TypeScript template for a router and does not involve unsafe execution methods.

agency-autonomous-optimization-architect