What problem does it solve? When building a multi-agent AI marketing platform, inconsistent agent implementations lead to unmaintainable code, untracked costs, and unpredictable behavior. This Skill enforces a single canonical pattern so every agent follows the same base class contract, prompt structure, I/O schema, and instrumentation rules. ## Core Features & Use Cases - Standardized Agent Structure: Every agent subclasses a common Agent base class with a single run(input) -> output method, nested Pydantic Input/Output models with strict config, and a four-section system prompt (Role, Constraints, Output Format, Anti-goals). - Brand Brain RAG Injection: Retrieves top-3 brand guideline chunks via pgvector cosine similarity and injects them as a <brand_context> XML block before every Anthropic call. - Cost and Observability Instrumentation: Writes AgentLog rows on run start and completion, tracking token usage and cost per call with a budget target under $0.05 per call. - Use Case: When adding a new content-generation agent to the marketing platform, apply this Skill to scaffold the class, prompt, Pydantic schemas, RAG injection, logging, and a mocked pytest suite in one consistent pass. ## Quick Start Create a new agent for generating Instagram captions following the project's canonical agent pattern with tests and cost logging.
