project-development

Task-model fit assessment determines whether an LLM-driven or traditional approach suits your needs. This skill provides a methodology to evaluate your specific task requirements, data constraints, and cost-performance trade-offs to guide that decision before architecture selection.

LLM batch pipelines follow a canonical five-stage pattern: acquire raw data, prepare and validate inputs, process through the LLM, parse structured outputs, and render results. This skill provides templates and design patterns to implement each stage efficiently.

Cost estimation requires modeling token consumption across your pipeline stages, input volumes, and model selection. This skill includes playbooks for forecasting token usage, calculating per-unit expenses, and planning resource allocation before deployment.

File-system state management works for structured batch pipelines when data flow is linear and idempotent. This skill covers when filesystem approaches suffice versus when databases become necessary based on pipeline complexity and recovery requirements.

Start with a quick manual prototype using sample data to test task-model fit before investing in pipeline infrastructure. This skill guides rapid prototyping, fit validation, and early architectural decisions to reduce wasted engineering effort.

Agent project structure must handle data flow, state persistence, and error recovery as complexity grows. This skill provides canonical pipeline patterns, resource planning, and validation methodology to design scalable agent architectures from the start.