Merge Systems: Unified FUZZ + AI Runner

The Big Picture

┌─────────────────────────────────────────────────────────────────────────────┐
│                         CURRENT STATE (3 Systems)                           │
└─────────────────────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────────────────┐
│                    express-repo-to-abi-data (Backend API)                   │
│  - Already has /claude/jobs routes!                                         │
│  - Launches ECS Fargate tasks                                               │
└─────────────────────────────────────────────────────────────────────────────┘
           │                                        │
           ▼                                        ▼
┌─────────────────────────┐          ┌─────────────────────────────────────┐
│      runner             │          │      recon-magic-framework          │
│  (TypeScript)           │          │  (Python)                           │
│  ubuntu:24.04           │          │  python:3.12-slim                   │
│  ~3GB image             │          │  ~1.8GB image                       │
│                         │          │                                     │
│  ECHIDNA, MEDUSA        │          │  CLAUDE_CODE, OPENCODE              │
│  HALMOS, FOUNDRY        │          │  PROGRAM                            │
└─────────────────────────┘          └─────────────────────────────────────┘


┌─────────────────────────────────────────────────────────────────────────────┐
│                         TARGET STATE (1 System)                             │
└─────────────────────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────────────────┐
│                    express-repo-to-abi-data (Backend API)                   │
│  - Same routes                                                              │
│  - Launches unified runner                                                  │
└─────────────────────────────────────────────────────────────────────────────┘
                                    │
                                    ▼
┌─────────────────────────────────────────────────────────────────────────────┐
│                         UNIFIED RECON RUNNER                                │
│  ubuntu:24.04 | ~3.5GB image                                                │
│                                                                             │
│  ┌───────────────────┐  ┌───────────────────┐  ┌───────────────────┐       │
│  │    FUZZ TOOLS     │  │    AI TOOLS       │  │   SHARED TOOLS    │       │
│  │  echidna, medusa  │  │  claude-code      │  │  foundry, node    │       │
│  │  halmos, slither  │  │  opencode-ai      │  │  python, go       │       │
│  │  bitwuzla, yices  │  │                   │  │  git, aws-cli     │       │
│  └───────────────────┘  └───────────────────┘  └───────────────────┘       │
│                                                                             │
│  Dispatch: job.type → runFuzzJob() OR runAIJob()                            │
└─────────────────────────────────────────────────────────────────────────────┘

Why Merge?

1. Backend Already Supports AI Jobs

express-repo-to-abi-data/src/routes/claude/jobs.ts    ← exists!
express-repo-to-abi-data/src/routes/claude/invites.ts ← exists!

2. 70% Tool Overlap

Tool	Runner	Magic	Both Need
Node 20	✅	✅	✅
Python	✅	✅	✅
Go	✅	✅	✅
Foundry	✅	✅	✅
Echidna	✅	✅	✅
Medusa	✅	✅	✅
Halmos	✅	❌	✅
claude-code	❌	✅	✅

3. Same Infrastructure Pattern

Both:

Clone a repo
Run commands in repo
Upload results
Update job status
Need monorepo support (foundry root detection)

4. Hybrid Jobs Become Possible

AI-guided fuzzing
Fuzzer results → AI analysis
AI writes handlers → Fuzzer tests them

Unified Dockerfile (Draft)

FROM ubuntu:24.04

# === SHARED BASE ===
RUN apt-get update && apt-get install -y \
    curl gcc make python3-pip python3-venv unzip jq wget tar \
    git build-essential cmake ninja-build ripgrep sudo \
    libboost-all-dev flex bison libffi-dev zlib1g-dev

# === NODE 20 ===
ENV NODE_VERSION=20.18.0
RUN curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.3/install.sh | bash
ENV NVM_DIR=/root/.nvm
ENV PATH="$NVM_DIR/versions/node/v${NODE_VERSION}/bin:${PATH}"
RUN . "$NVM_DIR/nvm.sh" && nvm install ${NODE_VERSION} && corepack enable

# === PYTHON ===
RUN python3 -m venv /opt/venv
ENV PATH="/opt/venv/bin:$PATH"
RUN pip3 install solc-select hexbytes slither-analyzer uv

# === GO ===
RUN wget https://go.dev/dl/go1.22.5.linux-amd64.tar.gz && \
    tar -C /usr/local -xzf go1.22.5.linux-amd64.tar.gz
ENV PATH="$PATH:/usr/local/go/bin"

# === FOUNDRY ===
RUN curl -L https://foundry.paradigm.xyz | bash && foundryup
ENV PATH="$PATH:/root/.foundry/bin"

# === FUZZ TOOLS ===
# Echidna (binary release, NOT homebrew)
RUN wget https://github.com/crytic/echidna/releases/download/v2.3.0-RC2/echidna-2.3.0-RC2-x86_64-linux.tar.gz && \
    tar -xf echidna-*.tar.gz && mv echidna /usr/bin/

# Medusa
RUN git clone https://github.com/crytic/medusa /tmp/medusa && \
    cd /tmp/medusa && go build && mv medusa /usr/local/bin/

# Halmos + Solvers
RUN uv tool install --python 3.12 halmos
# Bitwuzla, Yices (see full Dockerfile)

# === AI TOOLS ===
RUN npm install -g @anthropic-ai/claude-code opencode-ai

# === AWS CLI ===
RUN curl "https://awscli.amazonaws.com/awscli-exe-linux-x86_64.zip" -o "awscliv2.zip" && \
    unzip awscliv2.zip && ./aws/install

# === APP ===
WORKDIR /app
COPY . .
RUN yarn install
RUN pip install -e ./magic  # Magic framework as library

ENTRYPOINT ["yarn", "start"]

Unified Runner Logic

// src/index.ts
import { Job } from "@prisma/client";

async function main(job: Job) {
  switch (job.type) {
    // Fuzzing jobs (existing)
    case "ECHIDNA":
    case "MEDUSA":
    case "HALMOS":
    case "FOUNDRY":
      return runFuzzingJob(job);

    // AI jobs (new)
    case "AI_WORKFLOW":
      return runWorkflowJob(job);  // Uses magic framework
    case "AI_PROMPT":
      return runPromptJob(job);    // Direct claude/opencode call

    default:
      throw new Error(`Unknown job type: ${job.type}`);
  }
}

async function runWorkflowJob(job: Job) {
  // Clone repo
  await exec(`git clone ${job.repoUrl} repo`);

  // Detect foundry root (shared problem!)
  const foundryRoot = await detectFoundryRoot("repo");

  // Run magic framework
  await exec(`cd repo && python -m magic \
    --workflow ${job.workflowName} \
    --foundry-root ${foundryRoot}`);
}

Migration Steps

Phase 1: Dockerfile Merge

Create unified Dockerfile in runner/
Add AI tools (claude-code, opencode-ai)
Remove homebrew from magic (use echidna binary)
Test fuzzing jobs still work
Test AI jobs work

Phase 2: Runner Dispatch

Add job type dispatch logic
Import magic framework as library
Add runWorkflowJob() and runPromptJob()

Phase 3: Backend Updates

Ensure /claude/jobs creates jobs with correct type
Update starter.ts command generation
Add AI-specific env vars to ECS task

Phase 4: Deprecate Magic Worker

Remove standalone worker.py polling
Magic framework becomes library only
All jobs launch via ECS

Shared Problems (Solve Once)

1. Foundry Root Detection

def detect_foundry_root(repo_path):
    for root, dirs, files in os.walk(repo_path):
        dirs[:] = [d for d in dirs if d not in ['.git', 'node_modules', 'lib']]
        if 'foundry.toml' in files:
            return os.path.relpath(root, repo_path)
    return "."

2. Job Failure Status

await prisma.job.update({
  where: { id: job.id },
  data: {
    status: success ? "COMPLETED" : "FAILED",
    errorMessage: error?.message
  }
});

3. Monorepo Support

const path = job.directory !== "." ? `repo/${job.directory}` : "repo";
// Used by BOTH fuzzing and AI jobs

Benefits Summary

Before (Separate)	After (Merged)
2 Docker images	1 Docker image
2 build pipelines	1 build pipeline
2 ECS task definitions	1 task definition
Duplicate foundry/node/go	Shared tools
Can't mix AI + Fuzz	Hybrid jobs possible
2 codebases to maintain	1 unified runner

Image Size Comparison

Runner alone:           ~3.0 GB
Magic alone:            ~1.8 GB (homebrew bloat)
─────────────────────────────────
Separate total:         ~4.8 GB

Unified (optimized):    ~3.5 GB
─────────────────────────────────
Savings:                ~1.3 GB (27%)

Next Action

Start with Dockerfile merge - lowest risk, highest impact:

cd ../runner
# 1. Add AI tools to existing Dockerfile
# 2. Build and test fuzzing still works
# 3. Test AI tools work
# 4. Push unified image

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Merge Systems: Unified FUZZ + AI Runner

The Big Picture

Why Merge?

1. Backend Already Supports AI Jobs

2. 70% Tool Overlap

3. Same Infrastructure Pattern

4. Hybrid Jobs Become Possible

Unified Dockerfile (Draft)

Unified Runner Logic

Migration Steps

Phase 1: Dockerfile Merge

Phase 2: Runner Dispatch

Phase 3: Backend Updates

Phase 4: Deprecate Magic Worker

Shared Problems (Solve Once)

1. Foundry Root Detection

2. Job Failure Status

3. Monorepo Support

Benefits Summary

Image Size Comparison

Next Action

FilesExpand file tree

Merge Systems.md

Latest commit

History

Merge Systems.md

File metadata and controls

Merge Systems: Unified FUZZ + AI Runner

The Big Picture

Why Merge?

1. Backend Already Supports AI Jobs

2. 70% Tool Overlap

3. Same Infrastructure Pattern

4. Hybrid Jobs Become Possible

Unified Dockerfile (Draft)

Unified Runner Logic

Migration Steps

Phase 1: Dockerfile Merge

Phase 2: Runner Dispatch

Phase 3: Backend Updates

Phase 4: Deprecate Magic Worker

Shared Problems (Solve Once)

1. Foundry Root Detection

2. Job Failure Status

3. Monorepo Support

Benefits Summary

Image Size Comparison

Next Action