kodelyth-ecc 1.7.2 → 1.7.4

package/social/twitter-threads.md

@@ -0,0 +1,621 @@
+# Kodelyth ECC — Twitter / X Hype Content
+# v1.7.4 · github.com/sifxprime/kodelyth-ecc
+
+---
+
+## THREAD 1 — Devil Mode: 8 Adversarial Agents
+**Theme:** Red-team your code before attackers do
+**Image:** social/hype-devil-mode.svg
+
+---
+
+**Tweet 1/8**
+Most "security reviews" are vibes checks.
+
+Kodelyth ECC ships with 8 dedicated adversarial agents that read your code the way an attacker would.
+
+One command. Runs in parallel. 8 minutes.
+
+🧵 /devil-mode — what it actually does
+
+---
+
+**Tweet 2/8**
+`/devil-mode` fires 4 agents simultaneously:
+
+→ prompt-injection-hunter — finds AI injection vectors
+→ supply-chain-auditor — hunts malicious deps
+→ secret-hunter — scans for hardcoded credentials
+→ backdoor-hunter — detects inserted vulnerabilities
+
+All running in parallel. Not sequentially.
+
+---
+
+**Tweet 3/8**
+`/devil-mode --all` adds 4 more:
+
+→ jailbreak-tester — live AI feature red-team
+→ code-stealer-detector — code provenance audit
+→ license-violation-finder — GPL contamination check
+→ chaos-engineer — failure mode analysis
+
+8 agents. 8 minutes. Was 60+ doing this by hand.
+
+---
+
+**Tweet 4/8**
+The prompt-injection-hunter checks things most security tools miss entirely:
+
+- Indirect prompt injection in RAG pipelines
+- System prompt leaks via tool outputs
+- Tool hijacking in MCP servers
+- Agent-to-agent injection vectors
+
+AI features have a new attack surface. ECC maps it.
+
+---
+
+**Tweet 5/8**
+supply-chain-auditor goes deeper than `npm audit`:
+
+- Typosquatting detection
+- Dependency confusion vectors
+- Lockfile integrity verification
+- Install-script analysis
+- Recent CVE cross-reference
+
+Your `node_modules` is the largest third-party code surface you own. It should be audited.
+
+---
+
+**Tweet 6/8**
+Devil mode is non-destructive.
+
+It analyzes, reports, and stops.
+
+Nothing gets modified. Nothing gets sent anywhere. No cloud. No account. No API key.
+
+Just adversarial analysis, local, fast, and yours.
+
+---
+
+**Tweet 7/8**
+When to run `/devil-mode`:
+
+- Before going open-source: `--pre-public`
+- Before shipping to production: `--pre-launch`
+- After a security incident: `--all`
+- Any time you think like an attacker
+
+Semantic routing also auto-triggers it: "think like an attacker, sweep my code" works.
+
+---
+
+**Tweet 8/8**
+Kodelyth ECC is free. MIT. No account.
+
+npx kodelyth-ecc
+
+70 agents. 194 skills. 97 commands. 8 adversarial.
+11 platforms. 0 cloud.
+
+github.com/sifxprime/kodelyth-ecc
+
+[attach: social/hype-devil-mode.svg]
+
+---
+---
+
+## THREAD 2 — Compound Learning: Gets Smarter Every Session
+**Theme:** The toolkit that learns your style
+**Image:** social/hype-compound-learning.svg
+
+---
+
+**Tweet 1/8**
+Most AI tools forget everything the moment a session ends.
+
+Kodelyth ECC has a three-layer memory system that compounds across every session you run.
+
+Month 3 feels like a colleague who has worked with you for years.
+
+🧵 How compound learning actually works
+
+---
+
+**Tweet 2/8**
+Layer 1: Project Lessons
+
+Every correction you give Claude gets encoded permanently.
+
+Say "use pnpm not npm" once. It writes to tasks/lessons.md. The next session, it injects that rule before your first message.
+
+You never say it again.
+
+---
+
+**Tweet 3/8**
+The mechanics:
+
+Session ends → capture-correction.js scans for correction signals
+→ extracts the rule → appends to tasks/lessons.md
+
+Next session starts → read-lessons.js fires first
+→ injects "PROJECT LESSONS — HARD RULES" into context
+→ Claude follows them. Every time.
+
+12 correction signal patterns. Async. Zero latency.
+
+---
+
+**Tweet 4/8**
+Layer 2: Global Memory
+
+~/.kodelyth/memory/ stores solutions cross-project via BM25 search.
+
+Every prompt you type triggers a recall. If you solved a similar problem in a different project six months ago, that solution surfaces before the AI responds.
+
+No cloud. No embeddings. Pure local.
+
+---
+
+**Tweet 5/8**
+Layer 3: Semantic Intent Routing
+
+70 specialist agents route from intent, not keywords.
+
+Paste a stack trace → debug-detective
+Paste code → code-reviewer
+"help me build X" → /project-launch
+"production is down" → incident-commander
+
+No config. No agent names to memorize. Just describe the problem.
+
+---
+
+**Tweet 6/8**
+Three layers working together:
+
+Layer 1 knows your project rules.
+Layer 2 knows your past solutions.
+Layer 3 knows your intent.
+
+Combined: the ramp-up cost at session start drops to near zero. You start exactly where you left off.
+
+---
+
+**Tweet 7/8**
+This is all local.
+
+tasks/lessons.md — plain Markdown you own and edit
+~/.kodelyth/memory/ — plain JSONL you can inspect and delete
+Rules — just files in ~/.claude/rules/
+
+Nothing is transmitted. Nothing is analyzed on a server. No model is being fine-tuned on your corrections.
+
+---
+
+**Tweet 8/8**
+Kodelyth ECC. Free. MIT. No account.
+
+npx kodelyth-ecc
+
+70 agents. 11 platforms. Compound learning.
+194 skills. 97 commands. 0 cloud.
+
+github.com/sifxprime/kodelyth-ecc
+
+[attach: social/hype-compound-learning.svg]
+
+---
+---
+
+## THREAD 3 — Parallel Agents: The Speed Thread
+**Theme:** Fire 4 specialists simultaneously — not one after another
+**Image:** social/hype-parallel-agents.svg
+
+---
+
+**Tweet 1/8**
+A full code review used to mean:
+code quality → security → performance → API contracts
+
+Done sequentially. Each agent gets the full context. 60 minutes.
+
+/team-review fires all 4 simultaneously. 15 minutes. Same depth.
+
+🧵 How parallel agents work and why it matters
+
+---
+
+**Tweet 2/8**
+The 8 parallel commands:
+
+/team-review — 4 agents, 15 min
+/project-launch — 5 agents, 10 min
+/security-audit — 3 agents, 8 min
+/debug-blitz — 3 agents, 8 min
+/refactor-sprint — 4 agents, 12 min
+/pre-release — 3 agents, 8 min
+/onboard — 3 agents, 10 min
+/devil-mode — 4-8 agents, 8-15 min
+
+---
+
+**Tweet 3/8**
+/project-launch is the most underrated one.
+
+You describe a project idea. Five agents fire simultaneously:
+
+architect — system design
+pair-programmer — implementation approach
+security-reviewer — threat model
+tdd-guide — test strategy
+ux-reviewer — interaction and a11y
+
+Catch design mistakes before writing a single line.
+
+---
+
+**Tweet 4/8**
+Why parallel matters beyond speed:
+
+Each agent gets a focused, narrow context.
+They don't pollute each other's reasoning.
+You get independent opinions, not one AI trying to do everything.
+
+code-reviewer doesn't soften its findings because security-reviewer already complained.
+
+---
+
+**Tweet 5/8**
+/debug-blitz is for bugs that have resisted 30+ minutes of investigation.
+
+debug-detective + silent-failure-hunter + env-debugger.
+
+Three hypotheses in parallel:
+- It's a code bug
+- It's a silent failure / swallowed error
+- It's an environment / config difference
+
+One of them usually wins fast.
+
+---
+
+**Tweet 6/8**
+/pre-release is the "go/no-go" command.
+
+release-captain + security-reviewer + code-reviewer.
+
+Checks: semver decisions, changelog completeness, security regressions, code quality gates, rollback plan, smoke test coverage.
+
+Run it before you tag. Not after.
+
+---
+
+**Tweet 7/8**
+You don't need to memorize command names.
+
+Semantic routing auto-triggers parallel commands from intent:
+
+"review my project before I deploy" → /team-review
+"launching tomorrow, find everything" → /devil-mode --pre-launch
+"help me build a SaaS from scratch" → /project-launch
+
+Just describe what you need.
+
+---
+
+**Tweet 8/8**
+Kodelyth ECC. Free. MIT. No account.
+
+npx kodelyth-ecc
+
+70 agents. 8 parallel commands. 194 skills.
+11 platforms. Compound learning. 0 cloud.
+
+github.com/sifxprime/kodelyth-ecc
+
+[attach: social/hype-parallel-agents.svg]
+
+---
+---
+
+## THREAD 4 — MCP Server: Bridge Every AI Framework
+**Theme:** One toolkit, every AI tool you use
+**Image:** social/hype-mcp-server.svg
+
+---
+
+**Tweet 1/8**
+Claude Desktop, LangGraph, AutoGen, CrewAI, OpenAI Agents SDK.
+
+Every framework wants its own integration layer.
+
+Kodelyth ECC ships an MCP server with 16 tools, 6 prompts, and 377 resources.
+
+One server. Any client.
+
+🧵 The kodelyth-ecc MCP server
+
+---
+
+**Tweet 2/8**
+npx kodelyth-ecc mcp
+
+That's it. stdio JSON-RPC server starts locally.
+
+Connect it to Claude Desktop with two lines in settings.json.
+Connect it to LangGraph with the MCP client adapter.
+Connect it to any framework that speaks Model Context Protocol.
+
+---
+
+**Tweet 3/8**
+The 16 tools cover the full ECC surface:
+
+agent_invoke — call any of the 70 specialist agents
+memory_search — BM25 recall from your local memory store
+skill_load — load any of 194 skills into context
+swarm_run — spawn parallel agents in git worktrees
+cost_route — get model-tier recommendation for a task
+session_analyze — analyze session transcripts
+
++10 more.
+
+---
+
+**Tweet 4/8**
+377 resources means every agent, skill, and command is directly addressable as an MCP resource.
+
+Your LangGraph workflow can reference kodelyth://agents/debug-detective and get the full agent spec.
+
+Your AutoGen setup can load kodelyth://skills/tdd-guide dynamically.
+
+---
+
+**Tweet 5/8**
+The 6 system prompts:
+
+- Intent routing setup
+- Memory protocol
+- Security analysis
+- Code architecture
+- Testing strategy
+- Incident response
+
+Pre-built, battle-tested system prompts you can load into any AI agent without writing them yourself.
+
+---
+
+**Tweet 6/8**
+Localhost only. No external network access.
+
+The MCP server listens on stdio or a configurable port. Never exposes anything to the internet. All operations are read-only from disk (except writing to tasks/lessons.md when you ask it to).
+
+No API key. No account. Runs entirely on your machine.
+
+---
+
+**Tweet 7/8**
+The dashboard (npx kodelyth-ecc dashboard) complements the MCP server:
+
+5 tabs: Overview · Memory · Evolve · Catalog · Sessions
+
+Browse all 70 agents and 194 skills in a browser UI.
+View and manage your memory store.
+See session transcripts and token usage.
+
+Also localhost only. Also free.
+
+---
+
+**Tweet 8/8**
+Kodelyth ECC. Free. MIT. No account.
+
+npx kodelyth-ecc
+npx kodelyth-ecc mcp
+npx kodelyth-ecc dashboard
+
+70 agents. 16 MCP tools. 377 resources.
+11 platforms. 0 cloud.
+
+github.com/sifxprime/kodelyth-ecc
+
+[attach: social/hype-mcp-server.svg]
+
+---
+---
+
+## THREAD 5 — The Full Pitch: What ECC Actually Is
+**Theme:** The complete picture for someone who has never heard of it
+**Image:** social/hype-stats-hero.svg
+
+---
+
+**Tweet 1/8**
+Most "AI agent kits" are folders of markdown files you have to memorize.
+
+Kodelyth ECC is different. It has semantic intent routing baked into the rules layer.
+
+You describe your problem in plain words. The AI announces which specialist takes over. You get senior-grade help.
+
+🧵 What ECC actually is
+
+---
+
+**Tweet 2/8**
+The numbers:
+
+70 specialist agents (including 8 adversarial)
+194 skills
+97 slash commands (8 parallel multi-agent)
+22+ automation hooks
+14 always-on rules
+11 platforms, 13 install targets
+336 passing tests
+
+One install. npx kodelyth-ecc
+
+---
+
+**Tweet 3/8**
+The semantic routing system is the core feature.
+
+It reads intent, not keywords:
+
+"this thing is broken" → debug-detective
+"I have no idea where to start" → kodelyth-advisor
+"production is throwing 500s" → incident-commander
+[paste code, no text] → code-reviewer
+[paste stack trace, no text] → debug-detective
+
+You never type an agent name unless you want to.
+
+---
+
+**Tweet 4/8**
+The 11 platforms:
+
+Claude Code (full: 70 agents, 194 skills, 97 commands, 22+ hooks)
+Windsurf · Cursor · Codex CLI · Cline · Roo Code
+Aider · Kimi · Gemini CLI · Antigravity · OpenCode
+
+Feature depth varies by platform. Full power on Claude Code.
+Agents + skills + rules on most others.
+
+---
+
+**Tweet 5/8**
+The specialist agent catalog is deep:
+
+debug-detective — evidence-first root cause
+incident-commander — P0/P1 production triage
+load-tester — k6, Locust, capacity planning
+image-architect — Gemini/DALL-E/fal.ai/SVG generation
+kodelyth-memory — local BM25 self-learning
+migration-guide — framework version upgrades
+release-captain — semver, tagging, rollback plan
+
+...and 63 more.
+
+---
+
+**Tweet 6/8**
+The compound learning system means you configure it once:
+
+Tell it "use pnpm" once → it never uses npm again.
+Tell it "no comments on trivial lines" once → it learns your style.
+Month 3: it works like someone who has been on your team for years.
+
+Entirely local. Zero cloud.
+
+---
+
+**Tweet 7/8**
+What you get at install:
+
+~/.claude/agents/ — 70 specialist agent files
+~/.claude/commands/ — 97 slash commands
+~/.claude/skills/ — 194 skill files
+~/.claude/hooks/ — 22+ automation hooks
+~/.claude/rules/ — 14 always-on rules
+
+All editable. All yours. No lock-in. MIT license.
+
+---
+
+**Tweet 8/8**
+Kodelyth ECC. Free. MIT. No account. No API key. No telemetry.
+
+npx kodelyth-ecc
+
+That's the entire install.
+
+github.com/sifxprime/kodelyth-ecc
+
+[attach: social/hype-stats-hero.svg]
+
+---
+---
+
+## STANDALONE HOOK TWEETS (5)
+
+---
+
+**Hook 1 — Devil Mode**
+Your AI reviewed your code.
+
+An attacker is about to review it too.
+
+/devil-mode fires 8 adversarial agents that read code the way an attacker does. Prompt injection. Supply chain. Secrets. Backdoors. Jailbreaks. All parallel. 8 minutes.
+
+npx kodelyth-ecc
+github.com/sifxprime/kodelyth-ecc
+
+[attach: social/hype-devil-mode.svg]
+
+---
+
+**Hook 2 — Speed**
+You've been doing code reviews wrong.
+
+Sequentially: code quality → security → performance → API contracts. 60 minutes.
+
+/team-review fires all 4 specialists simultaneously. 15 minutes. Same depth.
+
+70 agents. 8 parallel commands. Free forever.
+
+npx kodelyth-ecc
+
+[attach: social/hype-parallel-agents.svg]
+
+---
+
+**Hook 3 — Compound Learning**
+You corrected your AI assistant once: "use pnpm not npm."
+
+It forgot by the next session.
+
+Kodelyth ECC encodes every correction to disk. Next session, it injects that rule before your first message.
+
+Month 3: it works like someone who has been on your team for years.
+
+npx kodelyth-ecc — free, local, no cloud.
+
+[attach: social/hype-compound-learning.svg]
+
+---
+
+**Hook 4 — Stats**
+70 specialist agents.
+194 skills.
+97 slash commands.
+22+ automation hooks.
+11 platforms.
+336 passing tests.
+0 cloud.
+
+One install: npx kodelyth-ecc
+
+MIT. No account. No API key. No telemetry.
+
+github.com/sifxprime/kodelyth-ecc
+
+[attach: social/hype-stats-hero.svg]
+
+---
+
+**Hook 5 — MCP**
+npx kodelyth-ecc mcp
+
+That starts an MCP server with 16 tools, 6 system prompts, and 377 resources.
+
+Connect Claude Desktop. LangGraph. AutoGen. CrewAI. OpenAI Agents SDK.
+
+Localhost only. No API key. Free forever.
+
+github.com/sifxprime/kodelyth-ecc
+
+[attach: social/hype-mcp-server.svg]

package/social/x-card-agents-grid.svg

@@ -69,7 +69,7 @@
   <rect x="480" y="280" width="360" height="120" rx="8" fill="#0d2b1d" stroke="#059669" stroke-opacity="0.4" stroke-width="1"/>
   <rect x="480" y="280" width="4" height="120" rx="2" fill="url(#mem)"/>
   <rect x="646" y="286" width="78" height="18" rx="9" fill="#059669" fill-opacity="0.25" stroke="#059669" stroke-opacity="0.5" stroke-width="1"/>
-  <text x="685" y="299" font-size="9" fill="#10b981" text-anchor="middle" font-weight="700">NEW v1.5.2</text>
+  <text x="685" y="299" font-size="9" fill="#10b981" text-anchor="middle" font-weight="700">v1.7.4</text>
   <text x="500" y="306" font-size="13" font-weight="700" fill="#ffffff">image-architect</text>
   <text x="500" y="325" font-size="12" fill="#8b949e">"my site looks plain / boring"</text>
   <text x="500" y="341" font-size="12" fill="#8b949e">"I need an OG image / social card"</text>
@@ -103,6 +103,6 @@
   <text x="920" y="424" font-size="32" font-weight="800" fill="#ffffff" text-anchor="middle">12</text>
   <text x="920" y="446" font-size="10" fill="#484f58" text-anchor="middle" letter-spacing="2">PLATFORMS</text>
   <line x1="860" y1="464" x2="980" y2="464" stroke="#21262d" stroke-width="1"/>
-  <text x="920" y="496" font-size="20" font-weight="800" fill="#ffffff" text-anchor="middle">v1.7.0</text>
+  <text x="920" y="496" font-size="20" font-weight="800" fill="#ffffff" text-anchor="middle">v1.7.4</text>
   <text x="920" y="516" font-size="10" fill="#484f58" text-anchor="middle" letter-spacing="2">LATEST</text>
 </svg>

package/social/x-card-free.svg

@@ -51,9 +51,9 @@
   <text x="800" y="446" font-size="14" fill="#8b949e">Every session → better context</text>
   <text x="800" y="476" font-size="12" fill="#484f58">Month 1: knows your style.</text>
   <text x="800" y="494" font-size="12" fill="#484f58">Month 3: works like it's been here years.</text>
-  <text x="800" y="528" font-size="11" fill="#30363d">Compound learning — v1.5.1</text>
+  <text x="800" y="528" font-size="11" fill="#30363d">Compound learning — v1.7.4</text>
 
   <!-- Footer CTA -->
   <text x="80" y="606" font-size="20" fill="#484f58">$ npx kodelyth-ecc</text>
-  <text x="1120" y="606" font-size="16" fill="#484f58" text-anchor="end">v1.5.8 · github.com/sifxprime/kodelyth-ecc</text>
+  <text x="1120" y="606" font-size="16" fill="#484f58" text-anchor="end">v1.7.4 · github.com/sifxprime/kodelyth-ecc</text>
 </svg>

package/social/x-card-hook.svg

@@ -30,25 +30,25 @@
   <rect x="80" y="252" width="3" height="68" rx="1" fill="url(#mem)"/>
   <text x="96" y="276" font-size="11" font-weight="700" fill="#10b981" letter-spacing="1">SESSION START — read-lessons.js</text>
   <text x="96" y="296" font-size="13" fill="#8b949e">Reads tasks/lessons.md → injects hard rules + project DNA</text>
-  <text x="96" y="311" font-size="11" fill="#484f58">before your first message · fires first · v1.5.1</text>
+  <text x="96" y="311" font-size="11" fill="#484f58">before your first message · fires first · v1.7.4</text>
 
   <rect x="80" y="336" width="490" height="68" rx="8" fill="#0d1a0d" stroke="#059669" stroke-opacity="0.3" stroke-width="1"/>
   <rect x="80" y="336" width="3" height="68" rx="1" fill="url(#mem)"/>
   <text x="96" y="360" font-size="11" font-weight="700" fill="#10b981" letter-spacing="1">STOP — capture-correction.js</text>
   <text x="96" y="380" font-size="13" fill="#8b949e">Scans session JSONL for corrections → writes to lessons.md</text>
-  <text x="96" y="395" font-size="11" fill="#484f58">12 signal patterns · async · zero latency · v1.5.1</text>
+  <text x="96" y="395" font-size="11" fill="#484f58">12 signal patterns · async · zero latency · v1.7.4</text>
 
   <rect x="80" y="420" width="490" height="68" rx="8" fill="#161b22" stroke="#21262d" stroke-width="1"/>
   <rect x="80" y="420" width="3" height="68" rx="1" fill="#a78bfa"/>
   <text x="96" y="444" font-size="11" font-weight="700" fill="#a78bfa" letter-spacing="1">SESSION START — memory inject</text>
   <text x="96" y="464" font-size="13" fill="#8b949e">BM25 recall from ~/.kodelyth/memory/ → past solutions</text>
-  <text x="96" y="479" font-size="11" fill="#484f58">cache-friendly prefix · auto-fires · v1.4.0</text>
+  <text x="96" y="479" font-size="11" fill="#484f58">cache-friendly prefix · auto-fires · v1.7.4</text>
 
   <rect x="80" y="504" width="490" height="68" rx="8" fill="#161b22" stroke="#21262d" stroke-width="1"/>
   <rect x="80" y="504" width="3" height="68" rx="1" fill="#a78bfa"/>
   <text x="96" y="528" font-size="11" font-weight="700" fill="#a78bfa" letter-spacing="1">USER PROMPT — auto chat recall</text>
   <text x="96" y="548" font-size="13" fill="#8b949e">Watches every prompt → injects relevant memories before AI responds</text>
-  <text x="96" y="563" font-size="11" fill="#484f58">per-session dedup · skips trivial prompts · v1.4.1</text>
+  <text x="96" y="563" font-size="11" fill="#484f58">per-session dedup · skips trivial prompts · v1.7.4</text>
 
   <!-- Right column — quality hooks -->
   <rect x="606" y="252" width="514" height="68" rx="8" fill="#161b22" stroke="#21262d" stroke-width="1"/>
@@ -78,5 +78,5 @@
   <!-- Footer -->
   <text x="80" y="618" font-size="16" fill="#484f58">All hooks fire automatically after</text>
   <text x="340" y="618" font-size="16" fill="#10b981" font-weight="600"> npx kodelyth-ecc</text>
-  <text x="1120" y="618" font-size="14" fill="#484f58" text-anchor="end">v1.5.8</text>
+  <text x="1120" y="618" font-size="14" fill="#484f58" text-anchor="end">v1.7.4</text>
 </svg>