Skip to content

hyperfocus 1.0.0

Latest
Latest

Choose a tag to compare

View all tags
@TEJASVISJAIN TEJASVISJAIN released this 03 Oct 18:12
· 6 commits to main since this release
v1.0.0
3de31e6

Questions on time.

  • A project brief, learnt in the background. When hyperfocus starts in a project it writes a short
    brief from your README, CLAUDE.md or AGENTS.md, build manifests, the folder tree, recent commit
    subjects and the files you've changed (each size-capped, secrets redacted, sensitive files never
    read), and caches it until HEAD moves. Every question call includes it.
  • The first question at your prompt, about the plan for it, before the agent edits anything. Each
    such question names the prompt it is about.
  • Question calls run in an empty folder, so they no longer load your project's CLAUDE.md or the
    agent's memory (about 1,300 tokens per call, and the cause of questions about "your memory notes").
  • Streamed: the first question shows as soon as it is written.
  • A warm next call: the next question process is started early and waits for its prompt, so
    Claude Code's start-up is paid in the background. If it isn't needed it is closed without calling
    the model, so it costs nothing.
  • The terminal opens the quiz once a question is ready, not only after an edit; a watching VS Code
    panel gets each question the moment it exists.
  • --stats shows time to first question: the median wait after a prompt and how many runs had a
    question before the agent finished. Measured and kept on your machine.

Questions worth answering.

  • Shorter: one idea, two lines at most, options of a few words. Longer ones are dropped.
  • Always sourced: a file and lines, or the plan for your prompt; a question with neither is dropped.
  • Spaced repeats: a missed question comes back after 1 day, then 3 and 7 days as you get it right,
    at most one per batch beside new questions, and only while its code is still there.
  • A lesson after a miss: press e for a short explanation of the idea, with your own code as the
    example. Follow-ups now stay on the question's topic.
  • Local model writer: --writer ollama (or "writer": "ollama") writes questions with a model on
    your machine through Ollama; --doctor checks that it runs and the model is pulled.

Ready for everyone.

  • The bridge protocol, the config file and the data files are versioned, documented in
    docs/FORMATS.md, and only grow within 1.x. New entries carry "v": 1.
  • Platform support is stated plainly: macOS, Linux and Claude Code are supported; Windows, Codex CLI,
    Gemini CLI and Open VSX editors are experimental.

Install: npm install -g @ddalus/hyperfocus · VS Code: Marketplace, or install the attached .vsix.

Assets 3
Loading