Supercharge-AI

Multi-agent framework for Claude Code. Moves context to markdown, delegates recursively to workers, and learns from its mistakes through persistent memory.

Inspired by Recursive Language Models (RLM) and Confucius Code Agent (CCA) — recursive self-delegation from RLM, persistent note-taking and hierarchical orchestration from CCA.

Installation

From marketplace (stable)

claude plugin marketplace add ac8ai/Supercharge-AI
claude plugin install supercharge-ai
supercharge init --add-permissions

From local clone (beta)

git clone https://github.com/ac8ai/Supercharge-AI.git
cd Supercharge-AI && git checkout beta
supercharge init --add-permissions

Local installs auto-pull the latest beta and install editable on each session start — no manual updating needed.

What happens

On first session start, the plugin auto-installs uv and the supercharge CLI if missing. Prompts are injected automatically via hooks. supercharge init adds the SuperchargeAI include to your project's CLAUDE.md. The --add-permissions flag adds permission entries so you don't get constant approval dialogs.

Linux and macOS only.

How it works

Once installed, SuperchargeAI takes over Claude Code's delegation. Instead of doing everything in one context window, it:

1. Plans — decomposes your request into scoped tasks
2. Delegates — sends each task to a specialized agent (code, review, research, etc.)
3. Recurses — agents can spawn workers, workers can spawn sub-workers (up to 5 levels deep)
4. Remembers — harvests learnings into persistent memory after each task

Everything flows through markdown files in .claude/SuperchargeAI/ — task briefs, notes, results, memory. Context windows are temporary; the markdown is permanent.

Agent types

Agent	Purpose
plan	Decompose requests into structured task lists
code	Implement features, fix bugs, write tests
document	Update documentation to reflect changes
research	Search the web, gather external context
review	Code review of completed work
consistency	Check for contradictions, broken references, duplication
memory	Maintain project and methodology memory from task results

Key commands

supercharge init              # Set up SuperchargeAI in your project
supercharge deinit            # Remove it
supercharge dashboard         # Web UI for metrics and session traces
supercharge version           # Check installed version

Most other commands (task init, subtask init, memory run, etc.) are used by the agents themselves — you don't need to run them manually.

Configuration

Override in .claude/settings.json or .claude/settings.local.json:

{
  "env": {
    "SUPERCHARGE_MAX_RECURSION_DEPTH": "3",
    "SUPERCHARGE_MAX_TURNS": "50"
  }
}

Variable	Default	Purpose
SUPERCHARGE_MAX_RECURSION_DEPTH	5	How many levels deep workers can spawn sub-workers
SUPERCHARGE_MAX_TURNS	(none)	Limit worker turns per invocation
SUPERCHARGE_FAST_MODELS	haiku	Models that use fast (fire-and-forget) worker mode

Branching and releases

• main — stable releases, what marketplace users get
• beta — pre-release testing via local clone
• Feature branches merge into beta, then beta into main

Version format follows PEP 440: 0.4.0 (stable), 0.4.0b1 (beta). See .claude/CLAUDE.md for details.

Developer docs

• docs/stack-propagation.md — env var, context, and identifier flow through the orchestrator/agent/worker stack