bearishsun/context-compiler — explained in plain English
Analysis updated 2026-05-18
Plan and execute a large multi-file refactor while minimizing tokens spent by the most expensive AI model.
Have cheaper AI agents scout a codebase and record file locations before a stronger model writes the implementation plan.
Resume a partially completed coding task from its saved run folder instead of starting over.
| bearishsun/context-compiler | aclark4life/home-depot-crawl | aetheria-labs1/storefront-skills | |
|---|---|---|---|
| Stars | 6 | 6 | 6 |
| Language | Python | Python | Python |
| Last pushed | — | 2014-08-10 | — |
| Maintenance | — | Dormant | — |
| Setup difficulty | moderate | moderate | easy |
| Complexity | 3/5 | 2/5 | 3/5 |
| Audience | developer | developer | pm founder |
Figures from each repo's GitHub metadata at analysis time.
Requires copying agent definitions into a Claude Code agents folder and configuring which models each stage uses.
Context Compiler is a skill for Claude Code, the AI coding assistant, that helps organize how a large coding task gets planned and carried out. Instead of letting one expensive AI model spend time reading through a whole codebase on its own, this project splits the work into four stages handled by different models, so the most capable and costly model only sees a short, focused document instead of the entire project. The four stages are explore, merge, plan, and implement. In the explore stage, one or more cheaper AI agents scan parts of the codebase and write down where relevant code lives, using file paths and line numbers rather than copying the code itself. In the merge stage, another agent combines those notes and runs a Python script that copies the actual code from disk into a single briefing document, keeping everything accurate since a script, not an AI, does the copying. In the plan stage, a stronger frontier model reads only that briefing document and writes a full step by step implementation plan with complete code changes. Finally, in the implement stage, a capable model executes the plan and produces a report of what it did. The idea behind this design is that AI models are most expensive when they have to keep re-reading everything they have already seen during back and forth exploration. By separating the scouting work from the actual planning and writing, the costly model does far less repetitive work. To use it, you copy the provided agent definitions into your Claude Code agents folder, either for one project or globally, and edit the planner agent's configuration to point at your strongest available model. The included Python script for merging code has no extra dependencies beyond Python itself. Each task gets its own run folder so multiple tasks do not interfere with each other, and a failed run can be resumed from where it left off. The project notes that this pipeline is meant for larger, multi-file tasks and is not worth using for small changes involving only one or two files.
Context Compiler is a Claude Code skill that splits large coding tasks across four AI stages so the expensive planning model reads a short briefing instead of exploring the whole repo.
Mainly Python. The stack also includes Python, Claude Code.
No license information was found in the README.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.