CesarMSFelipe/opencontext-runtime — reverse-engineered prompt

Build me an offline first tool for AI coding assistants that can prepare the right code context before the assistant starts working. I want it to index a local repo, trace the call graph, rank the most relevant symbols, stay inside a token budget, and return a verified context pack in one deterministic call instead of doing repeated search and full file reads.

It should feel like a practical command line app with a quick install, a demo mode I can run on my own repo, and an easy way to hook into editors and coding agents through MCP. I also want useful local tools around that, like search, explain, impact analysis, symbol aware edits, memory, and quality checks. Keep the retrieval path fully local and deterministic, add security defaults like redaction and secret scanning, and make it fail closed when needed.

If a model is available, support a guided software delivery loop with planning and implementation phases. If not, it should stay honest and output a structured plan only. Look up current docs online if you need to.