Show GN: Contexty — A tool that lets developers directly see and control the context AI sees

Contexty is an open-source tool that lets developers directly see and adjust the context AI coding agents reference.

The problem it addresses is clear. As context piles up, model performance drops, moving to a new chat means re-explaining premises, and /compact causes too much information to be lost. Above all, it's hard to check exactly which files and lines of code the AI is looking at when it reaches a conclusion.

The core features are as follows.

• HSCMM (Human-supervised Context Management)

  • Real-time tree visualization of the context the AI is currently referencing, in VSCode's Context Explorer panel
  • Can check included files, folders, and even specific lines of code
  • Unnecessary items can be removed instantly, and needed files or selected regions can be added directly
  • Lines included in the context are highlighted in the editor for visual distinction

• AASM (Active Agent-supervised Architecture)

  • A Sub-Agent that detects architectural anti-patterns at the prompt stage
  • Warns about or blocks mistakes like cramming all logic into one file, creating a God Object, or overusing global mutable state, before the AI generates code

• DCP (Dynamic Context Pruning)

  • Automatic compression before hitting the context limit
  • Transparently shows what was compressed and how many tokens were saved
  • Compressed parts can be restored via decompress if needed

• ACPM (Agent Context Permission Management)

  • Fine-grained control over the tools and file path permissions available to the AI
  • Permissions like file-read, file-write, shell, and web can be saved as per-project presets and reused

Installation starts with npx @ttalkkak-lab/opencode-contexty init. Once you select an IDE, the OpenCode extension and IDE extension are installed automatically, and most features are usable via the CLI alone.

The core message is simple: the more powerful AI coding tools become, the more important it is to have a mechanism for humans to understand and control what the AI is looking at. Contexty is a tool aiming to fill that gap.