Superdense
Superdense is a local developer tool that helps AI coding agents improve at real-world goals by running an outcome loop: it records what worked, ships the next move, and measures results. It solves the problem of agents starting cold each task, so good work never compounds. It is for developers using coding agents like Claude Code, Codex, Cursor, or OpenCode who want to systematically improve metrics like signups or latency.
✨ Key features
- Outcome loop with /outcome-setup, /outcome-run, /outcome-update commands
- Local SQLite store indexes sessions from multiple agents
- Session search and filtering by project, prompt, errors, tools, etc.
- Compactors provide salience and trace views of long sessions
- Hypothesis and experiment tracking with verdicts
- Cohort comparison of deliverables by run cost and outcomes
🎯 Use cases
- Improve signup conversion on a pricing page
- Grow qualified followers on a social media account
- Reduce p95 latency on a checkout API
- Carry context into new tasks with /chain
- Analyze past sessions for insights and recurring failures
📦 Installation
🧰 Requirements: Node.js 20+; works locally with agents like Claude Code, Codex, Cursor, OpenCode; no cloud required.
npm i -g @nimrobo/superdense # Node 20+
🚀 Usage
superdense studio # indexes your sessions, installs the skills, sets you up (UI at http://127.0.0.1:4242)
Then create a folder for each outcome, open your agent there, and run /outcome-setup to define the goal. From there, /outcome-run starts the loop and /outcome-update sharpens it over time.
❓ FAQ
What agents does Superdense work with?
It reads raw sessions from Claude Code, Codex, Cursor, and OpenCode via adapters.
Does Superdense require a cloud service?
No, it is local and nothing leaves your machine; it ships as one binary with indexer, server, and Studio UI.
How do I install the skills for my agent?
Run superdense skill install to install the outcome and session skills into Claude and Codex.
What is the purpose of the goal files?
They define the outcome: goal.md is the protected outcome, run.md is the working playbook, and gate.md is the completion contract.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.