dsj1984/agent-protocols

This is a collection of structured prompts and guidelines for steering LLM coding assistants—essentially a prompt engineering framework without code. Quantitative signals are extremely weak (4 stars, 0 forks, no velocity, 110 days old, no observable adoption or engagement). The approach is incremental: system prompts and persona-based instruction are well-established patterns in LLM engineering (OpenAI assistants, Anthropic's constitution-AI guidance, Claude prompt caching). The project has no novel technical contribution—it repackages known prompt-engineering techniques into a README. Integration surface is read-only documentation, not a composable component. Frontier risk is high because: (1) Anthropic, OpenAI, and Google already ship agent scaffolding, persona systems, and prompt libraries as first-class platform features; (2) adding better SDLC workflows to an agent interface requires zero additional infrastructure on their part; (3) the competitive advantage here is purely pedagogical/organizational, not technical. A frontier lab could trivially incorporate these ideas into their agent products. This lacks defensibility: no switching costs, no lock-in, no data gravity, no specialized tooling—just documentation that could be rewritten in an afternoon.