Kodelyth Ecc

[](https://www.npmjs.com/package/kodelyth-ecc) [](https://www.npmjs.com/package/kodelyth-ecc) [](https://github.com/sifxprime/kodelyth-ecc/stargazers) [](LICENSE)

Kodelyth ECC is a production-grade AI coding toolkit — 70 specialist agents (incl. an 8-agent devil-mode adversarial crew), 194 skills, 97 commands, a god-tier semantic intent-routing system, local self-learning memory, MCP server, swarm orchestrator, and an observability dashboard — all local, zero telemetry.

Works with Claude Code, Windsurf, Cursor, Codex CLI, Google Antigravity, OpenCode, Cline, Roo Code, Aider, Kimi, and Gemini CLI.

> No telemetry. No cloud. Just rules, agents, skills, MCP server, and your own private memory store — all on your disk. The local dashboard gives you full visibility without sending anything anywhere.

---

Why ECC ≠ Another Agent Collection

Most "AI agent kits" are folders of markdown files you have to remember the names of. ECC is infrastructure — a layered system where intent routing, compound memory, parallel orchestration, and quality hooks all reinforce each other.

You:   "I've been staring at this NullPointerException for two hours,
        I'm losing my mind."

AI:    → Routing to debug-detective (your error + frustration matches the bug-tracking signal)

       That kind of bug is exhausting — let's trace it properly so we
       fix the root cause, not the symptom.

       First, can you share the full stack trace and...

You never typed use debug-detective. You didn't have to. The toolkit read the intent, picked the specialist, and announced the routing. Next time you can invoke it directly — but you don't have to remember names to get senior-grade help.

The Layer Stack

| Layer | What it does | Other kits | |---|---|---| | Intent routing | Plain-language → right specialist via 10-tier priority rules | Mostly missing — you memorize names | | 70 agents | Specialists with playbooks, severity calibration, real commands | Often persona-only ("you are a senior engineer...") | | 194 skills | Domain knowledge files agents read on demand | Rarely separated from agents | | 97 commands | Slash workflows (/tdd, /devil-mode, /team-review) | Limited or none | | 8 parallel commands | Fire 3-8 agents simultaneously, aggregate results | Rare | | Compound memory | BM25 local recall + auto-inject + project lessons | Cloud-only or absent | | 22+ hooks | Quality gates, secret scan, project-DNA detection | Often missing | | 11 IDE platforms | Claude Code, Windsurf, Cursor, Codex, Antigravity, OpenCode, Cline, Roo Code, Aider, Kimi, Gemini CLI (13 install targets) | 1-2 platforms typical | | Zero telemetry | Everything stays on your disk; verifiable | Many kits phone home |

Quick Comparison vs Other Kits

| Feature | Kodelyth ECC | agency-agents | awesome-claude-agents | Generic prompt libs | |---|---|---|---|---| | Specialist agents | 70 | ~30 | ~20 | Varies | | Skills as separate layer | ✅ 194 | ❌ | ❌ | ❌ | | Slash commands | ✅ 97 | Some | Some | ❌ | | Parallel multi-agent commands | ✅ 8 (incl. /devil-mode) | ❌ | ❌ | ❌ | | Intent routing (plain language → agent) | ✅ 10-tier rule | ❌ | ❌ | ❌ | | Local BM25 self-learning memory | ✅ | ❌ | ❌ | ❌ | | Compound learning from corrections | ✅ tasks/lessons.md | ❌ | ❌ | ❌ | | Adversarial / red-team agents | ✅ 8 (devil-mode) | ❌ | ❌ | ❌ | | Quality hooks | ✅ 22+ | Some | ❌ | ❌ | | IDE platforms | 11 (Claude Code, Windsurf, Cursor, Codex, Antigravity, OpenCode, Cline, Roo Code, Aider, Kimi, Gemini CLI) | 1-2 | 1 | Varies | | Telemetry | ❌ none | Varies | ❌ | Varies | | Test coverage | ✅ 373 tests | ❌ | ❌ | ❌ | | Distributed via | npx, curl, clone | Manual | Manual | Manual |

---

Install

Option 1 — npx from npm (recommended, any platform)

npx kodelyth-ecc                              # Claude Code (default)
npx kodelyth-ecc --target windsurf-project    # Windsurf (per-project)
npx kodelyth-ecc --target windsurf-home       # Windsurf (global)
npx kodelyth-ecc --target cursor-project      # Cursor IDE
npx kodelyth-ecc --target codex-home          # Codex CLI
npx kodelyth-ecc --target antigravity         # Google Antigravity
npx kodelyth-ecc --target opencode            # OpenCode
npx kodelyth-ecc --target cline               # Cline (VS Code)
npx kodelyth-ecc --target roocode             # Roo Code (VS Code)
npx kodelyth-ecc --target aider               # Aider terminal agent
npx kodelyth-ecc --target kimi                # Kimi Code
npx kodelyth-ecc --target gemini-project      # Gemini CLI (project)
npx kodelyth-ecc --target gemini-home         # Gemini CLI (global)

Node.js 18+ required. Download Node if you don't have it.

Platform support at a glance

Feature depth varies by platform — hooks are a Claude Code native format, and some platforms have no agent/command concept:

| Platform | Agents | Skills | Commands | Hooks | Rules | |---|---|---|---|---|---| | Claude Code | ✓ 70 | ✓ 194 | ✓ 97 | ✓ 22+ | ✓ | | Roo Code | ✓ | ✓ | ✓ | — | ✓ | | Codex CLI | ✓ | ✓ | ✓ | — | ✓ | | Aider | ✓ | ✓ | ✓ | — | ✓ | | Kimi | ✓ | ✓ | ✓ | — | ✓ | | Windsurf | ✓ | ✓ | — | — | ✓ | | Antigravity | ✓ | partial | ✓ | — | ✓ | | Gemini CLI | ✓ | ✓ | — | — | ✓ | | Cursor | — | ✓ | — | — | ✓ | | Cline | ✓ | — | ✓ | — | ✓ | | OpenCode | — | — | — | — | ✓ |

Hooks use Claude Code's JSON settings format — no equivalent exists on other platforms. Cursor reads rules and skills from .cursor/; its agent system uses a different format not yet compatible with ECC agents.

Memory & dashboard reality per IDE

Memory storage is a single shared file at ~/.kodelyth/memory/memories.jsonl — every IDE on the same machine reads and writes the same memories. The only thing that varies is how memories surface:

| Platform | Auto-recall on every prompt | Auto-capture on success | Manual recall via MCP tool | Dashboard "Live IDE activity" | |---|---|---|---|---| | Claude Code | ✓ (hook) | ✓ (hook) | ✓ | ✓ Claude session files | | Windsurf | — | — | ✓ recall_memory | ✓ Windsurf + Windsurf-Next state | | Cursor | — | — | ✓ recall_memory | ✓ workspace storage dirs | | Codex CLI | — | — | ✓ recall_memory | — | | Antigravity | — | — | ✓ recall_memory | ✓ .agent/ in cwd | | Roo Code / Aider / Kimi / Cline / Gemini CLI / OpenCode | — | — | ✓ if MCP-capable | — |

What this means in practice:

• A memory captured in Claude Code is recall-able from every other IDE the same day — the file is shared.
• In Windsurf / Cursor / Codex / Antigravity, the AI does NOT auto-fire memory recall; the rules/common/memory-protocol.md rule (installed automatically) tells the AI to call the recall_memory MCP tool proactively at the start of substantive prompts.
• The dashboard's Sessions → Live IDE activity tab surfaces session files for Claude Code, Windsurf, Windsurf-Next, Cursor, and Antigravity. Add custom paths via the KODELYTH_EXTRA_IDE_WATCH env var (comma-separated).

# Watch additional paths in the dashboard
export KODELYTH_EXTRA_IDE_WATCH="$HOME/my-agent-logs,$HOME/other-tool/state"
npx kodelyth-ecc dashboard

Option 2 — npx from GitHub (always latest commit)

npx github:sifxprime/kodelyth-ecc

Same --target flags work.

Option 3 — curl (macOS / Linux only)

curl -fsSL https://raw.githubusercontent.com/sifxprime/kodelyth-ecc/main/install.sh | bash

With a target:

curl -fsSL https://raw.githubusercontent.com/sifxprime/kodelyth-ecc/main/install.sh | bash -s -- --target windsurf-project

Power Bundles

Pre-configured for who you actually are:

npx kodelyth-ecc --bundle indie-hacker    # Solo founder / SaaS — ship fast, validate, harden
npx kodelyth-ecc --bundle red-team        # Security engineer — devil-mode + adversarial workflows
npx kodelyth-ecc --bundle enterprise      # Compliance / audit team — SBOM, license, supply chain

Each bundle installs the full ECC toolkit (all 70 agents, 194 skills, 97 commands, 22+ hooks), adds a BUNDLE.md cheat sheet, and biases the AI toward audience-fit workflows on every session.

Combine with any target:

npx kodelyth-ecc --bundle red-team --target windsurf-project
npx kodelyth-ecc --bundle enterprise --target codex-home
npx kodelyth-ecc --bundle indie-hacker --target antigravity

Option 4 — Clone and run

git clone https://github.com/sifxprime/kodelyth-ecc.git
cd kodelyth-ecc

# macOS / Linux
./install.sh                              # Claude Code (default)
./install.sh --target windsurf-project    # Windsurf

# Windows (PowerShell)
.\install.ps1
.\install.ps1 -Target windsurf-project

---

MCP Server — Universal Adapter

Run ECC as a Model Context Protocol server and consume it from Claude Desktop, LangGraph, AutoGen, CrewAI, OpenAI Agents SDK, Cursor, Windsurf — anything that speaks MCP.

npx kodelyth-ecc mcp                         # stdio JSON-RPC server

What it exposes (all local, zero telemetry):

• 16 tools — route_intent, recall_memory, capture_memory, list_agents, get_skill, audit_skill_match, …
• 6 prompts — full intent routing rule, agents/skills/commands overviews, handoff chains, devil-mode
• 365 resources — every agent, skill, command, rule, and bundle addressable via kodelyth://... URIs

Drop into Claude Desktop in 30 seconds:

// claude_desktop_config.json
{
  "mcpServers": {
    "kodelyth-ecc": {
      "command": "npx",
      "args": ["-y", "kodelyth-ecc", "mcp"]
    }
  }
}

Full reference: [docs/mcp.md](docs/mcp.md).

---

How Intent Routing Works

The toolkit ships with a single rule file (rules/common/agent-intent-routing.md) that the AI loads automatically on every session. It maps what you say to the right specialist agent across 10 priority tiers.

Two activation paths

1. Explicit — type it directly:

use debug-detective
@code-reviewer
invoke security-reviewer

2. Implicit — just describe your problem; the AI routes you:

| What you write | Auto-routed to | |---|---| | "I'm stuck, no idea where to start" | kodelyth-advisor | | "I've been debugging this for hours" | debug-detective | | "nothing works, driving me crazy" | debug-detective | | "Should I use Context or Zustand?" | pair-programmer | | "help me build a todo app" | /project-launch | | "I have this idea for a SaaS dashboard" | /project-launch | | "I'm starting a new side project" | /project-launch | | "can you review my code?" | code-reviewer or /team-review | | "review my project before I deploy" | /team-review | | "is my project ready to ship?" | /team-review | | "my site looks plain, needs visuals" | image-architect | | "I need an OG image for my app" | image-architect | | "remember we always use pnpm here" | /lessons | | "Build failed on Vercel" | build-error-resolver | | "Is this JWT signing secure?" | security-reviewer | | "Why is this so slow?" | performance-optimizer | | "Plan the v2 migration" | planner + migration-guide | | "Tests pass locally but fail on CI" | flake-hunter + env-debugger | | "I lost my commits after reset --hard" | git-rescue | | "npm install is failing" | dependency-doctor | | "Cut a 1.4 release" | release-captain | | "Add accessibility to this form" | ux-reviewer | | "Open-source this project" | opensource-forker (chain) | | [paste code with no text] | code-reviewer | | [paste stack trace with no text] | debug-detective |

The AI always announces which agent is taking over (→ Routing to ) and always teaches you the explicit form for next time (Tip: type "use "). No silent personality changes.

Multi-agent chains

Real problems span multiple specialties. ECC ships standard handoff chains:

pair-programmer  →  tdd-guide  →  code-reviewer  →  security-reviewer
(approach)         (write tests)  (review impl)     (auth, validation)

debug-detective  →  tdd-guide        →  refactor-cleaner
(root cause)        (regression test)   (cleanup)

opensource-forker  →  opensource-sanitizer  →  opensource-packager  →  release-captain
(clean fork)          (strip secrets)            (README, license)        (cut v0.1.0)

See skills/agent-handoff/SKILL.md for the full handoff protocol and standard chains.

---

Parallel Agents — 8 Commands, Minutes Not Hours

Eight commands fire multiple specialist agents simultaneously and aggregate their results into a single structured report.

| Command | Agents Fired | Time Saved | |---|---|---| | /project-launch | architect + pair-programmer + security-reviewer + tdd-guide + ux-reviewer | 45 min → 10 min | | /team-review | code-reviewer + security-reviewer + performance-optimizer + api-guardian | 60 min → 15 min | | /security-audit | security-reviewer + dependency-doctor + api-guardian | 30 min → 8 min | | /debug-blitz | debug-detective + silent-failure-hunter + env-debugger | 60 min → 15 min | | /refactor-sprint | refactor-cleaner + code-simplifier + type-design-analyzer + tdd-guide | 45 min → 12 min | | /pre-release | release-captain + security-reviewer + code-reviewer | 30 min → 8 min | | /onboard | code-explorer + architect + doc-updater | 45 min → 12 min | | /devil-mode | 8 adversarial agents (see below) | Hours → 20 min |

Each command waits for all agents to complete, then returns a single Team Review Report with findings bucketed by severity: CRITICAL → HIGH → MEDIUM → LOW.

---

Devil Mode — 8 Adversarial Agents

8 adversarial agents that read your codebase the way an attacker would. Fire them in parallel with /devil-mode:

/devil-mode --pre-public    # before going open-source — full secret/license/IP sweep
/devil-mode --pre-launch    # before launch — adds AI red-team + chaos planning
/devil-mode --all           # all 8 adversarial agents in parallel

The crew: prompt-injection-hunter, supply-chain-auditor, secret-hunter, license-violation-finder, jailbreak-tester, code-stealer-detector, backdoor-hunter, chaos-engineer. Each ships with real bash-grep patterns, severity calibration, and remediation playbooks — not theatrical personas.

---

Live Dashboard — Full Visibility, Localhost Only

You have a working, real-time dashboard. One command launches it:

npx kodelyth-ecc dashboard
# Opens http://127.0.0.1:5747 in your browser

What's inside:

| Tab | What you see | |---|---| | Overview | Agent count, memory stats, session count, recent activity | | Memory | Browse, search, and manage your local BM25 memory store | | Evolve | Self-improving memory — review AI-proposed refinements | | Catalog | Full searchable index of all 70 agents, 194 skills, 97 commands | | Sessions | Live IDE activity (Claude Code, Windsurf, Windsurf-Next, Cursor, Antigravity) + orchestration/swarm sessions |

Real-time:

• SSE push every 3 seconds, only when at least one browser tab is connected (zero CPU otherwise)
• Watches memory writes, evolve signals, token-budget changes, and IDE session files
• "Last activity" auto-updates without a page refresh
• Set KODELYTH_EXTRA_IDE_WATCH=path1,path2 to watch additional paths

Security design:

• GET-only — no write endpoints accessible from the UI
• Localhost-bound — refuses connections from non-localhost Host headers (DNS rebinding protection)
• Hardened headers — Content-Security-Policy, X-Frame-Options, X-Content-Type-Options on every response
• Zero telemetry — no external network calls ever leave your machine
• Max 10 SSE clients — connection cap prevents resource exhaustion

Full reference: [docs/dashboard.md](docs/dashboard.md).

---

Kodelyth Memory — Local Self-Learning

The first time you s

…

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: sifxprime
• Source: sifxprime/kodelyth-ecc
• License: MIT
• Homepage: https://www.npmjs.com/package/kodelyth-ecc

Install and usage instructions live in the source repository linked above.