Main Guidelines

• The instructions, tools, prompts, etc. should be model / AI-tool agnostic.
  We want anyone to be able to use them, with any tool or model, both because different developers have different preferences, and for future-proofing.
• Context and tools are much more important than which model or AI tool we're using. We should include as much context and tools for agents as possible, about our workflows and repositories, so that they can work autonomously and won't need additional context from users.
• Automation - Everything should be automated / easily ran. We don't want agents to get stuck or require manual intervention.
• Modular design - Bloat and token-count are real issues. We should give specific agents, specific context. Information that isn't relevant with all agents, shouldn't be shared with all agents.
• Token-efficiency - Often overlooked, but the higher the token count, the more "confused" and less "sharp" LLMs get (and there are also higher costs). All docs, instructions, etc. should be concise and direct. We shouldn't include redundant / obvious / implied information in prompts or instructions.
• We should respect existing instructions / docs for specific repositories (e.g., "builder") when working on them, and not override them.

Repository Link:

https://gitlab.cee.redhat.com/myochpaz/aipcc-ai-workspace