Command Palette

Search for a command to run...

UnylyUnyly
Весь каталог

Kodelyth Ecc

БесплатноНе проверен

Production-grade AI coding toolkit — 70 agents (incl. devil-mode adversarial crew), 194 skills, 97 commands, parallel multi-agent commands, semantic intent rout

GitHubEmbed

Описание

Production-grade AI coding toolkit — 70 agents (incl. devil-mode adversarial crew), 194 skills, 97 commands, parallel multi-agent commands, semantic intent routing, self-learning memory, and a built-in MCP server (16 tools / 6 prompts / 377 resources) tha

README

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.

Now bundled with:

  • RTK (Rust Token Killer, Apache-2.0) — auto-compresses shell command output for 60-90% input token savings
  • Terse mode (ECC-native, inspired by Caveman) — 4-level output-token compression dial for 40-70% output savings
  • Codebase graph (via codebase-memory-mcp, MIT) — AST-parsed knowledge graph across 158 languages; structural queries at 99% fewer tokens than file-by-file grep
  • Interactive CLI — type kodelythecc alone in a terminal for an arrow-key menu with live update check, dashboard, background daemon

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

Install — One command, 11 platforms, any OS

The lazy install (one command, everything wired up)

npm i -g kodelyth-ecc
kodelythecc --target claude-code --codebase-graph
# then restart your AI tool

That's it. This single flow:

  1. Installs both binaries (kodelyth-ecc and short-form kodelythecc) to your PATH
  2. Copies 70 agents + 194 skills + 97 commands + 22 hooks + 14 rules into your AI tool's config dir
  3. Auto-installs RTK binary and wires its PreToolUse hook (input compression starts on next AI restart)
  4. Installs Terse mode skill + /terse and /terse-compress slash commands (dormant — user types /terse to activate)
  5. Auto-installs codebase-memory-mcp and registers its MCP entries in your AI tool (with --codebase-graph)
  6. Runs legacy-memory migration if you had ~/.kodelyth/ from an older install → ~/.kodelythecc/

After it finishes, run kodelythecc alone to open the interactive menu.

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

MCP Server — universal adapter for any agent framework

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 toolsroute_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.


How Intent Routing Works

Intent routing — plain language to the right specialist agent

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 <agent>) and always teaches you the explicit form for next time (Tip: type "use <agent>"). 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

Parallel agents — fire multiple specialists simultaneously

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

Devil Mode — 8 adversarial agents, red-team your own code

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

Dashboard — localhost observability, zero telemetry

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 headersContent-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.


RTK — Input Token Savings (60-90%)

RTK (Rust Token Killer) — Apache-2.0, single Rust binary — intercepts shell commands before they run and filters the output. Common commands like git status, ls, cargo test, docker ps return ~80% less text with zero information loss for the LLM.

ECC bundles RTK end-to-end:

  • Auto-installs the binary via Homebrew (macOS) or curl script (Linux / WSL)
  • Auto-configures the PreToolUse hook in your AI tool's settings
  • Live ledger surfaced in the dashboard's Token Savings tab

Manage from the CLI

kodelythecc rtk install                # install rtk binary
kodelythecc rtk enable --target X      # wire into one IDE
kodelythecc rtk enable --all           # wire into every ECC-installed IDE at once
kodelythecc rtk status                 # version + active integrations
kodelythecc rtk gain --all             # raw rtk savings output
kodelythecc rtk --help                 # focused help

Live proof (from this repo maintainer's Mac)

total_commands:    1,285
total_input:       7,964,612    (raw)
total_output:      2,858,501    (after RTK filter)
total_saved:       5,107,394    ← 64.1% average reduction

Terse Mode — Output Token Savings (40-70%)

RTK compresses input. Terse mode compresses output. Together they stack — savings on both sides of every turn.

ECC's own implementation, inspired by Caveman (MIT), independently written. Ships as a skill + two slash commands + a deterministic zero-dep memory-file compressor.

Four levels

Type /terse in your AI tool. Level sticks until you switch or session ends.

Level Style
/terse off Normal AI voice
/terse lite Light trim, drop filler ("basically", "essentially")
/terse full Telegram-style fragments (default)
/terse ultra Maximum compression, symbols over words

Byte-preserved always: fenced code blocks, inline code, shell commands, error text, URLs, file paths, identifiers, numbers, versions.

Compress memory files permanently

/terse-compress (or kodelythecc terse compress <file>) rewrites CLAUDE.md-style memory files into terse form so they cost fewer tokens every session forever. ~30% average byte reduction on real prose, 100% code/URL/path integrity.

CLI

kodelythecc terse status                # skill install state
kodelythecc terse stats                 # tokens saved from ledger
kodelythecc terse compress <file>       # rewrite a file, byte-preserves code
kodelythecc terse enable --all          # install skill + commands into every ECC IDE
kodelythecc terse --help

Codebase Graph — 158 Languages, Structural Queries

Powered by DeusData/codebase-memory-mcp (MIT). AST-parsed knowledge graph via tree-sitter across 158 languages, Hybrid LSP semantic type resolution for 11 major languages, cross-service HTTP/gRPC/GraphQL linking, 14 MCP tools for structural queries.

Why this matters: "Who calls ProcessOrder?" via file-by-file grep = ~412k tokens. Same question via the graph = ~3.4k tokens. That's a 99% reduction on structural questions.

Auto-install path

Pass --codebase-graph on ECC install and it auto-runs their official curl script + auto-configures the MCP server entry in every installed AI-coding agent.

kodelythecc --target claude-code --codebase-graph
# ↑ installs ECC + RTK + Terse + codebase-memory-mcp, all wired up

Then in your AI tool, say "Index this project". Done.

CLI

kodelythecc codebase install                              # install + auto-register
kodelythecc codebase status                               # version + indexed projects
kodelythecc codebase query search_graph '{"name_pattern": ".*Handler.*"}'
kodelythecc codebase query trace_path '{"function_name": "main"}'
kodelythecc codebase query get_architecture '{}'
kodelythecc codebase --help

Interactive CLI Menu

Type kodelythecc alone in a real terminal → arrow-key menu opens.

⚙ Kodelyth ECC  v2.3.0  ·  Elite Code Crew   up to date

 ▸ Open Dashboard                     localhost — RTK, Terse, Codebase, Memory
   Install ECC for another IDE        13-target picker
   RTK status                         Version + wired IDEs + savings
   Terse status                       Skill install state, ledger totals
   Codebase graph status              158 languages, structural queries
   Memory stats                       BM25 recall — captures, projects, tags
   Run in background                  Detached dashboard daemon
   Exit

↑/↓ navigate  ·  ⏎ select  ·  q / esc / Ctrl+C to quit

Behaviour rules

Situation Menu opens?
kodelythecc in Terminal / iTerm Yes
kodelythecc rtk status (any subcommand) No — runs subcommand
echo hi | kodelythecc (piped) No — runs installer
CI environment ($CI set) No
KODELYTH_NO_MENU=1 kodelythecc No

Uninstall

The menu's Uninstall ECC completely row runs an interactive full cleanup: removes the 759 ECC-installed files from ~/.claude/, unwires RTK from your AI tool, removes codebase-memory-mcp agent configs, removes ECC's MCP entry from Claude Code + Claude Desktop, and deletes ~/.kodelythecc/ (memory + ledgers). Prompts confirm before anything is deleted; a dry-run mode previews what would be removed without touching anything.

You can also run it non-interactively:

kodelythecc uninstall --dry-run           # preview what would be removed
kodelythecc uninstall --yes                # full cleanup
kodelythecc uninstall --yes --keep-memory  # remove files but keep ~/.kodelythecc/
npm uninstall -g kodelyth-ecc              # finally remove the npm package itself

Update check

The menu polls https://registry.npmjs.org/kodelyth-ecc/latest on open. Cached 24h in ~/.kodelythecc/update-check.json. When a newer version exists, an extra menu row appears at the top:

 ▸ Update to v2.3.1 [NEW]              npm i -g kodelyth-ecc

Background daemon

Selecting "Run in background" forks the dashboard as a detached process:

✓ Dashboard daemon started (pid 12345)
  URL:    http://127.0.0.1:5747
  Log:    /Users/you/.kodelythecc/dashboard-daemon.log
  Pid:    /Users/you/.kodelythecc/dashboard-daemon.pid
  Stop:   kill $(cat ~/.kodelythecc/dashboard-daemon.pid)

Survives shell exit. Real daemon.


Kodelyth Memory — Local Self-Learning

Kodelyth Memory — local BM25 self-learning memory

The first time you solve a hard problem, ECC remembers what worked. The next time you hit something similar, your AI surfaces the past solution before you ask.

Session 1 (March):
  You:  "Stripe webhook signatures are failing in production"
  AI:   [helps you debug, you discover raw body parser is required]
  You:  "Perfect, that worked, thanks"
  → ECC's Stop hook queues the lesson for review.
  → You confirm with /memory review-pending → stored locally.

Session 2 (August, new project):
  You:  "I need to add Stripe webhooks to checkout"
  AI:   I checked your memory — you solved this exact problem in March.
        Raw body parser before signature validation, test with stripe-cli
        not curl. Want me to apply the same pattern?
  You:  "Yes, do it"
  → 30 minutes of past debugging saved in 3 seconds.

How it works

Layer Mechanism
Capture A Stop hook scans your session JSONL, extracts (problem, approach, gotchas, tags). Queues for your review — never auto-stores.
Storage ~/.kodelyth/memory/memories.jsonl — append-only log on your disk only. Override with KODELYTH_MEMORY_DIR.
Retrieval BM25 keyword + tag search. Pure JS, sub-millisecond, no embeddings, no network.
Session-start injection A SessionStart hook builds a cache-friendly context block: stable prefix (your patterns + recent project memories) → variable suffix (relevant to current task).
Auto chat detection A UserPromptSubmit hook watches every message you type, runs BM25 search on it, and injects relevant memories before the AI responds. Per-session repeat suppression.
Cost win The stable prefix sits in the prompt cache. Anthropic charges 10% on cached tokens (5-min TTL); OpenAI auto-caches prefixes ≥1024 tokens. Long sessions become dramatically cheaper.

Privacy

Every byte stays on your machine. Verify any time with ls -la ~/.kodelyth/memory/. Sync across machines is opt-in (Dropbox/iCloud/git on that folder).

Slash command

/memory                          # Stats and recent memories
/memory recall <query>           # BM25 search
/memory remember "<title>"       # Capture (interactive — confirms before storing)
/memory review-pending           # Review Stop-hook candidates
/memory forget <id>              # Soft-delete a memory
/memory inject [--query <text>]  # Print what your AI sees about you

Honest limits

  • It is not model fine-tuning. The LLM never changes. We give it smarter context.
  • Cache savings apply to Anthropic and OpenAI. Other models (Gemini, Llama, Mistral) get the recall quality without the discount.
  • Cloud-AI platforms (Windsurf, Antigravity, partial Cursor) store sessions server-side. Auto-extract from past sessions doesn't work there. Manual /memory remember still does.

See skills/kodelyth-memory/SKILL.md for the full design + CLI reference.


Compound Learning System — Self-Improvement Loop

Compound learning — every correction stacks, Claude matches your style

Every correction you give Claude gets encoded into your project permanently. The toolkit gets smarter every session without any effort from you.

Session 1:  You type "use pnpm not npm"
            → Session ends
            → capture-correction.js scans the JSONL
            → Writes to tasks/lessons.md: "- use pnpm not npm"

Session 2:  read-lessons.js fires at session start
            → Injects: "PROJECT LESSONS — HARD RULES: - use pnpm not npm"
            → Claude uses pnpm without being told

Month 1:    10+ corrections stacked
            → Claude knows your naming style, preferred patterns, tech opinions
            → Zero ramp-up time on any new task

Month 3:    You try another tool
            → It uses npm. It uses the wrong pattern. It asks basic questions.
            → You come back.

Three-layer compound memory architecture

Layer File Scope How it works
Project Lessons tasks/lessons.md Per-project Hard rules from your corrections. Injected at session start as mandatory context.
Global Memory ~/.kodelyth/memory/ Cross-project BM25 fuzzy recall of past solutions. Auto-fires on every prompt you type.
Intent Routing 70 agents Always-on Routes your message to the right specialist from the first word. No names needed.

How it works under the hood

capture-correction.js — Stop hook, runs async at session end:

  • Scans session JSONL for 12 correction signal patterns ("no don't", "use X instead", "stop doing Y", "we always", "wrong approach", etc.)
  • Extracts them as plain-language rules
  • Appends them to tasks/lessons.md with date grouping

read-lessons.js — SessionStart hook, fires first before any other hook:

  • Reads tasks/lessons.md and formats rules as PROJECT LESSONS — HARD RULES block
  • Detects project DNA automatically: Node.js + framework (Next.js, React, NestJS, etc.), Go, Rust, Python, Java/Gradle, package manager (pnpm/bun/yarn/npm), test runner
  • Surfaces open tasks/todo.md items into session context

tasks/lessons.md — your project's rulebook

Edit it freely. Add rules manually. Remove rules that no longer apply. It's a plain markdown file at tasks/lessons.md in your project root.

# Claude Lessons

Project: **my-app**

## 2026-05-06

- use pnpm not npm
- never add try/catch without logging the error first
- we use Zod for validation, not Yup
- component files go in src/components, not src/app

What's Inside

Component Count Description
Agents 70 Specialist subagents — reviewers, planners, debuggers, architects, devil-mode adversarial crew, incident-commander, load-tester, memory, image-architect
Skills 194 Domain knowledge — patterns, testing, security, DevOps, intent routing, memory
Commands 97 Slash command workflows (/tdd, /plan, /memory, /devil-mode, /doctor, /update, etc.)
Parallel commands 8 /devil-mode, /team-review, /security-audit, /debug-blitz, /refactor-sprint, /pre-release, /onboard, /project-launch
Hooks 22+ Quality gates, secret scanning, branch checks, memory inject + capture + correction + project DNA
Rules 14 Always-on coding standards + intent routing + memory protocol + self-improvement workflow
Memory local BM25-indexed personal memory at ~/.kodelyth/memory/ (zero deps)

Agent Arsenal

70 specialist agents — one for every situation
70 specialist agents — agent grid preview

Kodelyth Exclusives — The 16 Agents That Make ECC

Agent One-line job
kodelyth-advisor Master guide — picks the right tool when you don't know where to start
kodelyth-memory Curates your local memory — recalls past solutions, captures new ones
pair-programmer The engineer who sits next to you before you write the code
debug-detective Never guesses — traces every bug to root cause
silent-failure-hunter Finds bugs that don't throw errors
incident-commander Runs production incidents — triage, contain, postmortem. P0s only.
load-tester Load and stress testing — k6, Locust, Artillery, capacity planning
ux-reviewer Reviews UX behavior + WCAG 2.1 AA accessibility (never touches design)
api-guardian Detects breaking API changes before they ship
migration-guide Framework / language version upgrades, phase by phase
dependency-doctor npm/pip/cargo/maven hell — CVE triage, lockfile diagnosis
git-rescue Recovers from broken git states without destroying history
release-captain Owns the release ritual — semver, tagging, publishing, rollback
env-debugger "Works on my machine" hunter — env, config, secrets, layers
flake-hunter Stabilizes flaky tests — never adds blind retries
image-architect AI image generation — Gemini/DALL-E/fal.ai/SVG, platform-aware

Agent Categories

Category Agents
Guidance kodelyth-advisor, pair-programmer, planner, architect, code-architect, chief-of-staff, migration-guide
Code Review code-reviewer, typescript-reviewer, python-reviewer, go-reviewer, rust-reviewer, java-reviewer, kotlin-reviewer, cpp-reviewer, csharp-reviewer, flutter-reviewer, database-reviewer, healthcare-reviewer
Build Fixers build-error-resolver, go-build-resolver, rust-build-resolver, java-build-resolver, kotlin-build-resolver, cpp-build-resolver, dart-build-resolver, pytorch-build-resolver, dependency-doctor, env-debugger
Debugging debug-detective, silent-failure-hunter, flake-hunter
Incident & Load incident-commander, load-tester
Security & API security-reviewer, api-guardian
Performance performance-optimizer
Quality refactor-cleaner, code-simplifier, type-design-analyzer
Testing tdd-guide, e2e-runner, pr-test-analyzer, flake-hunter
Documentation doc-updater, docs-lookup, comment-analyzer
Release & Ops release-captain, git-rescue
Open Source opensource-forker, opensource-packager, opensource-sanitizer
Specialized seo-specialist, ux-reviewer, code-explorer

Hooks — Always Running in Background

22+ automated quality hooks — PreToolUse, PostToolUse, Stop

Once installed (Claude Code target), these hooks run automatically with zero configuration:

Hook What it does
Session start — lessons Reads tasks/lessons.md and injects your hard rules + project DNA as context
Session start — memory Loads relevant past solutions from global BM25 memory
Auto chat recall Watches every prompt, injects relevant memories before AI responds
Correction capture Detects when you correct Claude, encodes the rule to tasks/lessons.md
Pre-commit Catches console.log, secrets, bad commit messages
Quality gate Runs type checks and formatting after edits
Git push reminder Prompts review before pushing
Config protection Blocks weakening of linter/formatter configs
Desktop notify macOS notification when a long task finishes
MCP health check Validates MCP servers before calling them
Test reminder Prompts to write tests when code is edited without tests
Branch name check Blocks git branches that don't match naming convention

Usage After Install

Start here (new users)

/kodelyth-quickstart

Core workflows — explicit invocation

use kodelyth-advisor      # Not sure what to do? Start here
use pair-programmer       # Think through approach before writing code
use planner               # Plan a feature before writing code
use code-reviewer         # Review after writing code
use debug-detective       # Trace a bug to its root cause
use api-guardian          # Check API changes for breaking contracts
use security-reviewer     # Review security-sensitive code
use ux-reviewer           # Review frontend UX and accessibility
use migration-guide       # Upgrading a framework or language version
use tdd-guide             # Write tests the right way
use performance-optimizer # Diagnose and fix slowness
use dependency-doctor     # npm/pip/cargo dep hell
use git-rescue            # Broken git state, lost commits, bad rebase
use release-captain       # Cut a clean release with rollback plan
use env-debugger          # "Works on my machine" — env/config/secrets
use flake-hunter          # Stabilize flaky tests

Or just describe your problem

The intent router will route you to the right one. The AI announces who's taking over, helps you, and tells you the explicit name for next time.

Load language patterns (skills)

/typescript-patterns    # TypeScript + React + Next.js
/python-patterns        # Python best practices
/golang-patterns        # Go best practices
/postgres-patterns      # Database patterns
/docker-patterns        # Container patterns
/coding-standards       # Universal baseline

Advanced skills

/git-mastery            # Trunk-based dev, rebase, bisect, monorepos
/observability          # Structured logging, metrics, OpenTelemetry, SLOs
/smart-debug            # Systematic debugging framework
/intent-routing         # How the auto-routing system works
/agent-handoff          # Standard multi-agent handoff chains

What Gets Installed

Claude Code — ~/.claude/

Source Destination What it does
agents/ ~/.claude/agents/ All 70 subagents available globally
skills/ ~/.claude/skills/ All 194 skills loadable via commands
hooks/hooks.json ~/.claude/hooks/ Automated quality gates
rules/ ~/.claude/rules/ Always-on standards + intent routing
commands/ ~/.claude/commands/ Slash commands (/tdd, /plan, etc.)

Windsurf — .windsurf/ (project) or ~/.codeium/windsurf/ (global)

Source Destination Notes
agents/ .windsurf/agents/ Call with use <agent-name> in Cascade
skills/ .windsurf/skills/ Domain knowledge, loadable via chat
rules/ .windsurf/rules/ Flattened rule files
rules/common/ .windsurfrules Concatenated — Windsurf reads this automatically every session

Cursor — .cursor/ in project root

Source Destination
rules/ .cursor/rules/
skills/ .cursor/skills/

Codex CLI — ~/.codex/

Source Destination
agents/ ~/.codex/agents/
skills/ ~/.codex/skills/
commands/ ~/.codex/commands/
rules/ ~/.codex/rules/

Antigravity — .agent/ in project root

Source Destination
agents/ .agent/skills/
commands/ .agent/workflows/
rules/ .agent/rules/

OpenCode — .opencode/ in project root

Source Destination
rules/ .opencode/rules/

Install Profiles — Language Bundles

Works with any target:

npx kodelyth-ecc --profile nextjs                              # Claude Code
npx kodelyth-ecc --target windsurf-project --profile nextjs    # Windsurf
npx kodelyth-ecc --target antigravity --profile python-api     # Antigravity

Available profiles:

Profile Includes
nextjs TypeScript + React + Next.js rules
python-api Python + Django/FastAPI rules
fullstack TypeScript + Python + Go rules
mobile Kotlin + Swift rules
backend Go + Python + Java rules

Or specify languages directly:

npx kodelyth-ecc typescript python golang rust java kotlin php swift cpp dart ruby elixir

Multi-Platform Support

Platform Supported Install target What gets installed
Claude Code Full claude-home (default) Agents, skills, commands, hooks, rules
Windsurf Full windsurf-project / windsurf-home Agents, skills, rules, .windsurfrules
Cursor Full cursor-project Rules, skills
Codex CLI Full codex-home Agents, skills, commands, rules
Google Antigravity Full antigravity Agents → skills, commands → workflows, rules
OpenCode Rules only opencode Rules (agents + skills not yet supported by OpenCode)

OS support: macOS, Linux (install.sh), Windows (install.ps1), or any OS with Node.js 18+ (npx).


Privacy & Philosophy

ECC is 100% local files. No telemetry, no cloud, no account, no tracking. Everything is markdown your AI reads on every session.

  • No phoning home — install scripts copy files and exit. MCP server, dashboard, swarm, and replay all stay on localhost.
  • You own everything — fork the repo, edit any agent, write your own.
  • Verifiablenpx kodelyth-ecc verify checks your install against the sha256 manifest.

Contributing

We welcome new agents, skills, hooks, and intent routing patterns that meet the Kodelyth Standard — specialist personas, model-agnostic, production-grade examples.

When adding a new agent, also update rules/common/agent-intent-routing.md with your trigger patterns so the intent router knows when to call you.

See CONTRIBUTING.md for templates, checklists, and the full Kodelyth Standard.


Changelog

See CHANGELOG.md. v1.8.0 highlights:

  • Visual system overhaulgithub-social-preview.svg and og-image.svg fully redesigned with two-panel layout: brand left, 2×2 stat cards right, dot grid texture, amber accent
  • SVG overflow fixes — KODELYTH wordmark, bottom bar text, and stats numerals all corrected to stay within canvas bounds across all 31 social assets
  • Test count update — 373 tests across 25 test files (up from 348)
  • 4K PNG exports — all 31 social assets regenerated at 4× scale; fullhd exports removed

v1.7.3 highlights:

  • Social hype pack — 5 X/Twitter SVG cards + 5 thread scripts, full SVG version sync, GitHub issue templates, repo polish

v1.7.0 highlights:

  • MCP Server — 16 tools, 6 prompts, 377 resources; stdio JSON-RPC; zero telemetry; works with Claude Desktop, LangGraph, AutoGen, CrewAI, OpenAI Agents SDK
  • Dashboard hardening — DNS rebinding protection, shell injection fix (execFileSync), CSP on all responses, SSE client cap (max 10), TOCTOU fixes
  • Devil Mode — 8 adversarial agents, --pre-public / --pre-launch / --all flags, real bash-grep patterns and remediation playbooks
  • 336 tests passing across 25 test files (now 373 at v1.8.0)

v1.5.8 highlights:

  • zsh inline comment fixnpx kodelyth-ecc --target windsurf-home # comment no longer crashes.
  • Dynamic install counts — agent/skill/command counts computed from filesystem at install time.
  • Ruby + Elixir language rules — coding-style, patterns, testing, security, and hooks.
  • /doctor command — health check on your ECC install without leaving your IDE.
  • /update command — upgrade to the latest version automatically.
  • 5 new parallel commands (/security-audit, /pre-release, /debug-blitz, /refactor-sprint, /onboard).

v1.5.2 highlights:

  • Added: image-architect agent — platform-aware AI image generation (Gemini Imagen 3, DALL-E 3, fal.ai, SVG)
  • Added: /project-launch — 5 founding-team agents fire in parallel (45 min → 10 min)
  • Added: /team-review — 4 audit agents fire in parallel (60 min → 15 min)
  • Added: /lessons command — cross-platform lesson loading

v1.5.1 highlights:

  • Added: capture-correction.js Stop hook — auto-encodes corrections to tasks/lessons.md
  • Added: read-lessons.js SessionStart hook — injects lessons + project DNA before first message
  • Added: rules/common/self-improvement-workflow.md — 6-pattern self-improvement loop

v1.5.0 highlights:

  • Added: incident-commander — production incident response
  • Added: load-tester — load/stress testing with k6, Locust, Artillery

v1.4.0 highlights:

  • Added: kodelyth-memory — local self-learning memory (BM25, zero deps, cache-friendly)
  • Added: Auto chat detection hook, SessionStart injection, Stop hook queue

Author

Shofiqul Islam — Creator of Kodelyth ECC

Kodelyth ECC is designed and maintained by Shofiqul Islam — full-stack engineer and AI tooling builder.

Platform Link
GitHub @sifxprime
X / Twitter @sifxprime
Facebook facebook.com/sifxprime
Instagram @sifxprime
npm npmjs.com/package/kodelyth-ecc

npm GitHub License Node

Built with craft. Zero telemetry. All yours.

from github.com/sifxprime/kodelyth-ecc

Установить Kodelyth Ecc в Claude Desktop, Claude Code, Cursor

Рекомендуется · одна команда, все IDE
unyly install kodelyth-ecc

Ставит в Claude Desktop, Claude Code, Cursor и VS Code — сам разбирается с npx, uvx и сборкой из исходников.

Впервые? Поставь CLI: curl -fsSL https://unyly.org/install | sh

Или настроить вручную

Выполни в терминале:

claude mcp add kodelyth-ecc -- npx -y kodelyth-ecc

FAQ

Kodelyth Ecc MCP бесплатный?

Да, Kodelyth Ecc MCP бесплатный — установка в пару кликов через Unyly без оплаты.

Нужен ли API-ключ для Kodelyth Ecc?

Нет, Kodelyth Ecc работает без API-ключей и переменных окружения.

Kodelyth Ecc — hosted или self-hosted?

Self-hosted: сервер запускается локально на твоей машине командой из раздела установки.

Как установить Kodelyth Ecc в Claude Desktop, Claude Code или Cursor?

Открой Kodelyth Ecc на unyly.org, выбери вкладку своего клиента (Claude Desktop, Claude Code, Cursor) и нажми Install — конфиг сгенерируется автоматически, без правки JSON.

Похожие MCP

Compare Kodelyth Ecc with

Не уверен что выбрать?

Найди свой стек за 60 секунд

Автор?

Embed-бейдж для README

Похожее

Все в категории development