Scalpel — Surgical AI for Your Codebase
12-dimension diagnostic. Adaptive surgical teams. Zero side effects.
Installation
# Claude Code
git clone https://github.com/anupmaster/scalpel.git ~/.claude/skills/scalpel
# Codex CLI
git clone https://github.com/anupmaster/scalpel.git ~/.agents/skills/scalpel
# Any project (pendrive style)
cd your-project && git clone https://github.com/anupmaster/scalpel.git .scalpel && .scalpel/install.sh
Usage
> Hi Scalpel, start work
> scalpel scan # Diagnostic only, no changes
> scalpel skip recon # Jump to task execution
Standalone Scanner (Zero AI)
./src/scanner.sh # Pretty terminal output
./src/scanner.sh --json # Machine-readable
./src/scanner.sh --score-only # Just the number: "83"
What It Does
- Diagnose — Scans 12 dimensions: stack, architecture, git forensics, database, auth, infrastructure, tests, security, integrations, agent ecosystem, code quality, performance
- Consult — Delivers Codebase Vitals health score, asks targeted questions
- Assemble — Designs a custom surgical team calibrated to YOUR project
- Operate — Spawns parallel agents in isolated worktrees with file jurisdiction
- Close — Delivery report, session memory saved, optional regression monitoring
Features
- Standalone scanner — Pure bash, zero AI, zero tokens. Runs anywhere.
- 12-dimension diagnostic — Deeper than any linter. Reads git history, deleted files, abandoned branches.
- Adaptive teams — Not templates. Teams shaped by YOUR project's needs.
- File jurisdiction — Each agent owns specific files. Zero merge conflicts.
- Agent scoring — Quality enforcement. Score < 50 = agent terminated.
- Session memory — Remembers previous sessions. Gets smarter over time.
- Pendrive architecture — Plug in, work, unplug. Zero trace.
- 7 agent adapters — Claude Code, Codex, Gemini, Cursor, Windsurf, Aider, OpenCode.
- GitHub Action — Automated PR health checks.
- Extensible config — Custom scan dimensions, team roles, scoring rules.
Permissions
Read-only during diagnostic. Write access only during surgical team operation.
Never modifies your existing CLAUDE.md, agents, commands, or settings.
Never commits itself to your git history. Uninstall leaves zero trace.
1---2name: scalpel3description: Surgical AI for your codebase. Plugs into any project, performs a 12-dimension diagnostic scan (git forensics, architecture, tech debt, security, infrastructure), delivers a Codebase Vitals health score, then assembles and manages an adaptive AI surgical team calibrated to YOUR project. Pendrive architecture — plug in, work, unplug, zero trace. Also ships a standalone bash scanner (zero AI, zero tokens) for instant codebase health checks.4license: MIT5---67# Scalpel — Surgical AI for Your Codebase8912-dimension diagnostic. Adaptive surgical teams. Zero side effects.1011## Installation1213```bash14# Claude Code15git clone https://github.com/anupmaster/scalpel.git ~/.claude/skills/scalpel1617# Codex CLI18git clone https://github.com/anupmaster/scalpel.git ~/.agents/skills/scalpel1920# Any project (pendrive style)21cd your-project && git clone https://github.com/anupmaster/scalpel.git .scalpel && .scalpel/install.sh22```2324## Usage2526```27> Hi Scalpel, start work28> scalpel scan # Diagnostic only, no changes29> scalpel skip recon # Jump to task execution30```3132## Standalone Scanner (Zero AI)3334```bash35./src/scanner.sh # Pretty terminal output36./src/scanner.sh --json # Machine-readable37./src/scanner.sh --score-only # Just the number: "83"38```3940## What It Does41421. **Diagnose** — Scans 12 dimensions: stack, architecture, git forensics, database, auth, infrastructure, tests, security, integrations, agent ecosystem, code quality, performance432. **Consult** — Delivers Codebase Vitals health score, asks targeted questions443. **Assemble** — Designs a custom surgical team calibrated to YOUR project454. **Operate** — Spawns parallel agents in isolated worktrees with file jurisdiction465. **Close** — Delivery report, session memory saved, optional regression monitoring4748## Features4950- **Standalone scanner** — Pure bash, zero AI, zero tokens. Runs anywhere.51- **12-dimension diagnostic** — Deeper than any linter. Reads git history, deleted files, abandoned branches.52- **Adaptive teams** — Not templates. Teams shaped by YOUR project's needs.53- **File jurisdiction** — Each agent owns specific files. Zero merge conflicts.54- **Agent scoring** — Quality enforcement. Score < 50 = agent terminated.55- **Session memory** — Remembers previous sessions. Gets smarter over time.56- **Pendrive architecture** — Plug in, work, unplug. Zero trace.57- **7 agent adapters** — Claude Code, Codex, Gemini, Cursor, Windsurf, Aider, OpenCode.58- **GitHub Action** — Automated PR health checks.59- **Extensible config** — Custom scan dimensions, team roles, scoring rules.6061## Permissions6263Read-only during diagnostic. Write access only during surgical team operation.64Never modifies your existing CLAUDE.md, agents, commands, or settings.65Never commits itself to your git history. Uninstall leaves zero trace.
anupmaster