One Harness to rule them all.
One set of rules, skills, and agents for Claude Code, Cursor, Windsurf, Copilot, and Cline, so your AI coding tools follow the same process.
$ npx sauron-ai init
20 Runtimes. One Source of Truth.
Sauron synchronizes rules and skills across your AI coding tools. Hover or focus a runtime to see its target configuration file.
Shared core
One AGENTS.md and one .agents/skills/ folder, read by most runtimes, so your rules live in a single place.
Safe merge
Sauron never replaces your files. It writes inside sauron:start and sauron:end blocks, your JSON keys win, and every change is backed up.
Fellowship of 9 Sub-Agents
Nine specialized personas coordinate every stage of the coding workflow, from planning to chaos testing.
Gandalf
Master Planner & Strategy
- ·Decomposes feature requests into atomic, sequenced tasks.
- ·Owns the roadmap and delegates work across the Fellowship.
Aragorn
Principal Systems Architect
- ·Defines system boundaries, data flow, and module contracts.
- ·Reviews architectural decisions before implementation begins.
Legolas
Precision Linter & Bug Hunter
- ·Runs static analysis and enforces coding-standard compliance.
- ·Flags subtle logic and type errors before merge.
Gimli
AST Refactorer & Dead Code Slasher
- ·Refactors code via AST-level transformations for safety.
- ·Removes dead code paths and unused exports.
Boromir
Security Shield & Gatekeeper
- ·Audits dependencies, secrets, and attack surfaces.
- ·Gates merges that touch auth, payments, or production data.
Frodo
Ringbearer & Focused Atomic Implementer
- ·Implements single, well-scoped tasks handed down by Gandalf.
- ·Keeps diffs minimal, atomic, and reviewable.
Samwise
State Keeper & Git Reconciler
- ·Manages branches, commits, and merge reconciliation.
- ·Keeps working state clean and recoverable.
Merry
QA & TDD Runner
- ·Writes and runs unit, integration, and E2E test suites.
- ·Enforces test-first development discipline.
Pippin
Edge Case & Chaos Prober
- ·Probes boundary conditions, race conditions, and malformed input.
- ·Surfaces chaos scenarios before they reach production.
Caveman Mode: Fewer Words, Same Code
Pick a tier during sauron init. Agent replies get shorter, while code, paths, and commands stay verbatim. Example shown below.
Standard Verbose Agent
Sure! I'd be happy to help you with that.
Let me start by explaining what I'm going to do...
First, I'll read the file to understand the context,
then I will carefully apply the requested change.
Sauron Caveman Engine
Read file.py.
Patch line 42: fix null check.
Verify tests pass.
Done.
Interactive Skills Explorer
Search and filter 370+ skills across 9 engineering departments, then copy a skill's slash command.
48
WorkflowTurn any rough prompt, half-formed idea, or task description into a finished, ready-to-send prompt optimized for Opus 4.8 (with adaptive thinking) inside the...
accessibility-auditor
FrontendAudits UI components for WCAG 2.2 AA compliance, keyboard navigation, and screen reader support, then outputs a remediated refactor.
adapt-project
WorkflowAnalyze a new or existing project repository, select matching Sauron capability modules, skills, prompts, and shared context files, and adapt Sauron governan...
adhd-communication-adapter
WorkflowShapes responses for an ADHD reader.
agent-config-generator
WorkflowA production-grade prompt to automatically generate best-practice AGENTS.md and CLAUDE.md files based on the actual repository context.
agent-guard
SecurityPre-execution safety hook, destructive command interceptor, and credential shield for autonomous AI coding agents across Claude Code and Copilot CLI.
agent-introspection-debugging
WorkflowStructured self-debugging workflow for autonomous agents covering failure state capture, context pressure analysis, root-cause diagnosis, and contained recov...
agent-observability
AI EngineeringTraces agent turns, tool calls, token spend, and failures into per-developer dashboards with privacy-safe capture policy.
aida-landing-page-structure
FrontendCodex/GPT-optimized AIDA page structuring, seed-based design randomization, and GSAP ScrollTrigger skeletons.
analyze-linkedin-post-performance
WorkflowParse exported LinkedIn post metrics CSV datasets to identify top hook structures, format conversion rates, engagement outliers, and content topics to scale.
anchor-author-voice
WorkflowMaintain strict alignment with a calibrated author voice profile across extended writing sessions, preventing stylistic drift, generic AI defaults, and vocab...
angular-principles
FrontendAngular v20+ engineering rules covering standalone components, signals reactivity, zoneless change detection, and strict project setup.
Showing 1–12 of 375 skills
Frequently Asked Questions
What does Sauron do?
Sauron writes one set of rules, skills, and agents into the files your AI coding tools already read, so each tool follows the same process.
Will it overwrite my files?
No. Sauron writes inside its own marked blocks, merges JSON settings with your keys winning, and keeps a backup of every change in .sauron/backups/. You can preview a run with --dry-run.
Does the CLI send my code anywhere?
No. The CLI has no telemetry and makes no network requests of its own. Your AI coding tools handle your code under their own terms.
What do I need to run it?
Node.js 18 or newer. Run npx sauron-ai init or pnpm dlx sauron-ai init in your project, then pick your runtimes.
Is Sauron free?
Yes. It is open source, and you can use it under your choice of the MIT, Apache-2.0, or Unlicense terms.